This application claims the benefit of European Patent Application

EP16178577.9 filed July 8th, 2016.

The present invention relates to a computer implemented method for the identification and characterization of sequence variants in nucleic acids. In particular, this method is able to quickly and accurately identify most types of sequence genome variations with a potential association to a disease, that is, from single nucleotide substitutions to large structural variants. This method may have multiple and direct applications in genomics-based diagnosis, prognosis and therapy. The invention further relates to a computer program and to systems suitable for performing such a method. The computer program may be designed to be lock-less and scalable, thereby allowing for high performance

implementations on parallel execution environments such as specialized hardware accelerators.

BACKGROUND ART

The genetic basis of disease is increasingly becoming more accessible thanks to the emergence of the Next Generation Sequencing (NGS) platforms, which have extremely reduced the costs and increased the throughput of genomic sequencing. For the first time in history, personalized medicine is close to becoming a reality through the analysis of each patient's genome. A wide range of genome variation of cells and individuals has been identified to be the direct cause, or a predisposition to genetic diseases: from single nucleotide variants (SNVs if they are somatic, and SNPs if they are

polymorphic in the population), to structural variants (SVs), which can correspond to deletions, insertions, inversions, translocations and copy number variations (CNVs), ranging from a few nucleotides to large genomic regions, including complete chromosome arms. These variations can exist between patients and also emerge among cells of the same patient. The unveiling of changes in the genome is driving discoveries such as the

Philadelphia translocation between chromosomes 9 and 22, whose presence implies the development of chronic myelogenous leukemia (CML) and its identification allows the development and selection of last-generation therapies.

The ideal exploitation of genomic sequencing should involve the accurate identification of all variants, in order to derive a correct diagnosis and to select the best therapy. For clinical purposes, it is important that this computational process be carried out within an effective timeframe. But a simple sequencing experiment typically yields thousands of millions of reads per genome, which have to be stored and analysed. The task is severely hindered by a variety of factors such as PCR-amplification and sequencing errors, limitations intrinsically linked to the size of the reads, biases in the sequencing techniques employed, the inherent repetitive and dynamic nature of the genomic sequences, and others. As a consequence, the analysis of genomes with diagnostic and therapeutic purposes is still a great challenge, both in the design of efficient algorithms and at the level of computing performance.

Modern medicine will rely on the identification of genetic markers for precision diagnosis and for the application of more specific therapies. Cancer is one of the most active diagnostic and therapeutic areas where genetic analysis is being applied. Having access to all somatic variation accumulated in a tumor cell is now allowing the study of the genetic causes of the tumor and the development of new clinical protocols that are already starting to be applied in some clinical centres, and that will be soon a reality for all modern healthcare systems around the world. This is why, the identification of tumor variants is key in research and soon, also in medical care. The variants responsible for the origin and progression of tumors are currently searched using a common scheme that involves the sequencing of both tumor and normal genome samples of the same patient, and the subsequent scan and identification of the differences between them. Most of the available methods rely on an initial step, where all the normal and tumor sequence reads are aligned to a reference genome to then identify the changes present in the tumor compared with the normal and reference genomes. Despite these methods have provided a great number of disease-associated variation so far, they still entail intrinsic limitations associated to the need of a

prealignment to a reference genome, affecting their performance and accuracy. More precisely, the reference-based identification of somatic variation in cancer genomes has currently the following sources of errors and limitations: (i) the initial alignment step, on which all the methods rely, is time consuming and particularly error prone with the tumor reads that carry the sequence variation, which are the most relevant for the analysis. It has been proven that many of these reads that carry changes and differences in their sequence are difficult or even impossible to align to the reference unmutated genome. The absence and the misplacement of tumor reads in the final alignment drastically affect all existing downstream methods for variant searching and calling. Although a number of alternative methods exist, this alignment step is generally performed with the same program (Li, H., et.al. "Fast and accurate short read alignment with Burrows-Wheeler transform" Bioinformatics 2009, vol. 25, pp. 1754-1760), which implies that nearly all analyses done nowadays share the same type of errors derived from this mapping of reads, (ii) The usage of a reference genome also involves the interference with millions of inherited variants (germline, i.e. not somatic) that affect both, the accuracy at the level of read mapping and the actual identification of the target somatic fraction (normally comprising only between 2 and 10 thousand variants). A considerable number of these germline variants are then frequently mispredicted as somatic changes, increasing the rate of false positives and decreasing, consequently, the final reliability and applicability of the results.

On top of the limitations and errors inherent to the generation and

dependency of this initial alignment, the subsequent analysis, where somatic changes are finally identified, also implies a number of restrictions and complications. For example, despite the great deal of possibilities in terms of available methods, not a single one of them is able to identify a wide range of somatic variation, but instead, each is limited to the detection of a particular size and type of mutation (Medvedev P. "Computational methods for discovering structural variation with next-generation sequencing" Nat- Methods Suppl. 2009, vol. 6, S13-S20)There are programs that use this alignment to detect only SNVs and others that only identify SVs, among which, each one is able to detect a particular variant size. For instance, some methods identify insertions or deletions that comprise a few nucleotides (from 2 to a few dozens), others detect medium size SVs (from a few dozens of nucleotides to a few hundreds), and a small fraction of them, are designed for the identification of larger SVs (Ding, L, et.al. "Expanding the computational toolbox for mining cancer genomes" Nat. Rev. Genet. 2014, vol. 15, pp. 556- 570). As the detection complexity increases with the detection range, methods designed for the identification of large SVs are also more imprecise in defining the exact location in the genome and the type of change, which are often necessary for being able to derive the functional consequences of the mutation. These programs often only report regions where SVs might be located. In order to overcome these limitations, to date, alternative approximations have been developed. The term reference-free becoming more popular and has recently been used to describe a wide range of fundamentally different underlying strategies. For example, these methodologies are using

fundamentally different unrelated strategies covering, from the use of reference mapping plus assembly-based (Chen K. "TIGRA: A targeted

Iterative Graph Routing Assembler for Breakpoint Assembly" Genome Res. 2016, vol. 24, pp.310-317), de novo assembly (Zhuang, J. "Local sequence assembly reveals a high-resolution profile of somatic structural variations in 97 cancer genomes" Nucleic Acid Research 2015, vol. 43, pp.8146-8156), to suffix tree approximations (Moncunill V., et al., "Comprehensive

characterization of complex structural variations by directly comparing genome sequence reads" Nature Biotech. 2014, vol. 32, pp. 1 106-1 1 12). The first two examples are based on the end-joining of reads in the tumor and normal genomes in order to identify discordant patterns. Although these assembly can also suffer the mapping-derived limitations, they have other major limitations associated to the underlying mechanism of the assembly, mostly when using NGS reads, as the overlapping regions needed to extend over the read size are often too small to be position-specific in the genome. Among other reference-free approximations reported to date, it can be highlighted a suffix tree-based method (SMUFIN) that compares in a tree-like structure all tumor and normal reads, to then extract discordant branches as candidate positions for variation. Although this particular way of analysing reads may directly overcome many of the limitations mentioned, it still lacks possibilities for detecting non-human sequences and is limited by the size of the tree, which grows in memory demands as sequencing coverage grows. Additionally, and in contrast to the approach followed in the method of the present invention, suffix trees are data structures that inherently require locking access patterns to allow for concurrent updates to take place, therefore limiting the ability to efficiently implement these approaches in high performance parallel computing systems. These two fundamentally different approaches of analysing sequence reads have also limitations of scalability, as the design of the code has not considered alternative ways of adapting to specific and more efficient hardware architectures. In fact, all the limitations mentioned hinder the incorporation of this type of genomic analysis into identification of somatic mutations applied to the clinical practice, which calls for much faster and more accurate computational methods. In addition, current methods for somatic variant calling still miss an important fraction of large SVs, which are relevant for the diagnosis and treatment of diseases.

Similar approximations have also been extended to deal with other type of problems in molecular biology. For example, some reference free methods have been developed to quantify the abundance of RNA isoforms from RNA- seq data (Patro R., et. al. "Sailfish enables alignment-free isoform

quantification from RNA-seq reads using lightweight algorithms" Nature Biotech. 2014, vol. 32, pp. 462-464), or to identify evolutionary-driven substitutions in homozygosis, using de novo assembly of plant genomes

(Nordstrom K.J.V, et.al. "Mutation identification by direct comparison of whole-genome sequencing data from mutant and wild-type individuals using k-mers" Nature Biotech. 2013, vol. 31 , pp. 325-331 ). Clearly quick and robust comparative methods, able to detect all kinds of SVs differentiating two states (normal vs. pathological, undifferentiated vs.

differentiated, etc.) with high sensitivity, specificity, speed and scalability are still needed, as well as systems and computer programs suitable for performing such methods.

SUMMARY OF THE INVENTION In contrast to what is found in the prior art, inventors have come up with a computer-implemented method for identifying nucleic acid variants between two genomic states that does not depend on the alignment of reads of either state to a reference genome, or on the construction of sequence-based suffix trees. Using a different underlying mechanism, this method is, on its own, able to accurately identify all types of variants (heterozygous and

homozygous), from single nucleotide variations to large structural variants at base pair resolution with unprecedented performance at the level of variant detection and execution possibilities. Importantly, the method is not restricted to the identification of a certain type of variant (SNVs, insertions, inversions, etc.) nor does it only perform for variants of a certain size, as is the case for many of the methods found in the art. Because of its underlying principle, based on the use of k-mers and hashtables independently of a reference genome, all the limitations that apply to all other existing methods (outlined above) are overcome. This translates to a method that is not only more robust, but is also more thorough and much faster than the methods described to date. This invention entails a reference-free detection method that allows discovering homozygous and heterozygous variation in genomes using a polymorphic k-mer strategy consisting in the sequential sub-selection of read regions that will be compared in different ways to finally isolate variant- containing regions. One of the key elements of this computer-implemented method is the way k-mers are handled, as they are taken as "dynamic entities" by taking their stems and using the latter to explore for their inflections and partial inflections (see below). Compared to the other reference-free methods described above, the invention relies on a

fundamentally different way of dealing with the reads since, instead of constructing an assembly, or a complete suffix-tree with all normal and tumor reads, it uses a particular k-mer strategy (see below) to fish reads with potential variation and discards, in one pass, the vast majority of reads with no information. This allows inventors to quickly filter and retain a subset of reads representing all the variants that are now computationally easy to treat and analyse. Of note, the use of k-mers for direct comparison of genomes has only been explored in simple scenarios with a small scope, data, and requirements (see for instance Nordstrom K., ibid). In general, the use of k-mers has some limitations due to their strict nature, and the way k-mers are distributed in genomes, requiring large amounts of computing resources if the identification of unique features is sought. Inventors address these limitations by using a more flexible approach: polymorphic k-mers, which in addition to k-mers, also identify variations (inflections) of k-mers with similar patterns (stems, see below). Unlike regular, fixed-length k-mers, polymorphic k-mers enable the identification of unique features, and at the same time provide the means to gather and group related sequences even if they are not strictly the same. This element is key, as will be seen in the examples found below.

Thus, a first aspect of the present invention is a computer-implemented method for identifying nucleic acid variants between two genomic states comprising the steps of: A) Inputting 2 sets of nucleic acid reads, which are sequences retrieved from a nucleotide sequencing method, wherein the first set of reads corresponds to cells representing a first test state, and the second set of reads corresponds to cells representing a second control state; B) Filtering the reads, wherein the filtering comprises: B1 ) Keeping only the reads with at least a percentage X1 of their bases with a Phred quality score higher than 20, being X1 equal to or above 90%; B2) Splitting the reads with an undefined nucleotide, giving one sequence before, and one sequence after the undefined nucleotide, the latter being discarded; and B3) Discarding the sequence reads with less than X2 bases, wherein X2 is from 25 to 50;

C) Generating a hashtable structure comprising: C1 ) Generating a number of N-X2+1 new reads for each read of sequence length N, wherein the new N- X2+1 reads correspond to all k-mers with length X2 nucleotides; and C2) Building a hashtable structure, which comprises all the k-mers generated in step C1 ) and further comprises the number of times each k-mer is observed in the two sets of reads corresponding to first and second states.

D) Detecting candidate variants in the sequence between first state and second state, wherein a k-mer of the hashtable structure is taken as a candidate breakpoint, which represents a variant between the first and second states, if it fulfills all the following requirements: D1 ) At least one inflection based on a k-mer's stem must have at least X3 reads with the same variation between first and second states, being X3 at least 2; D2) The percentage of first state reads in second state reads is not over a threshold X4, to account for possible contamination of control state reads with test state reads, wherein X4 is at least 5%. E) Clustering and filtering test and control reads derived from all candidate breakpoints accepted in step D to build blocks, by carrying out the steps: E1 ) Retrieving reads which contain the stem of at least one k-mer that represents the candidate breakpoint selected in step D); E2) The reads of step E1 ) with at least X5 k-mer variants within a window of X6 nucleotides are taken as leading reads, wherein X5 is at least 7 and X6 is at least 10; E3) Reads whose k-mers share at least one stem with a leading read are merged to give a block; and E4) If the nucleic acid whose variant is being identified is a double stranded DNA, then both forward and reverse variants are taken into account when building the block. F) Aligning blocks taking their leading reads as a reference: F1 ) For each read in the block, take the leading read's stem and find the longest inflection or partial inflection between the read and the leading read. F2) Successively position each read so that its matching inflection or partial inflection is aligned against the leading read. The performance and speed of this method make it more suitable for clinical applications (such as genomic analysis of cancer cells) than the alternative solutions, which result complex and time-consuming. As it is shown in the data below, the method has a superior performance even to last-generation methods such as the one published in Moncunill {ibid), which is also reference-free.

A second aspect of the present invention is a computer program product comprising program instructions for causing a computer system to perform the computer-implemented method for identifying nucleic acid variants between two genomic states of the first aspect of the invention.

The computer program product may be embodied on a storage medium (for example, a CD-ROM, a DVD, a USB drive, on a computer memory or on a read-only memory) or carried on a carrier signal (for example, on an electrical or optical carrier signal). The computer program may be in the form of source code, object code, a code intermediate source and object code such as in partially compiled form, or in any other form suitable for use in the implementation of the processes according to the invention. The carrier may be any entity or device capable of carrying the computer program.

For example, the carrier may comprise a storage medium, such as a ROM, for example a CD ROM or a semiconductor ROM, or a magnetic recording medium, for example a floppy disc or hard disk. Further, the carrier may be a transmissible carrier such as an electrical or optical signal, which may be conveyed via electrical or optical cable or by radio or other means.

When the computer program is embodied in a signal that may be conveyed directly by a cable or other device or means, the carrier may be constituted by such cable or other device or means.

Alternatively, the carrier may be an integrated circuit in which the computer program is embedded, the integrated circuit being adapted for performing, or for use in the performance of, the relevant methods.

A third aspect of the invention is a system for identifying nucleic acid variants between two genomic states, the system comprising:

A) Computer/Electronic means for inputting 2 sets of nucleic acid reads, which are sequences retrieved from a nucleotide sequencing method, wherein the first set of reads corresponds to cells representing a first test state, and the second set of reads corresponds to cells representing a second control state;

B) Computer/Electronic means for filtering the reads, wherein the filtering comprises: B1 ) Keeping only the reads with at least a percentage X1 of their bases with a Phred quality score higher than 20, being X1 equal to or above

90%; B2) Splitting the reads with an undefined nucleotide, giving one sequence before, and one sequence after the undefined nucleotide, the latter being discarded; and B3) Discarding the sequence reads with less than X2 bases, wherein X2 is from 25 to 50; C) Computer/Electronic means for generating a hashtable structure comprising: C1 ) Generating a number of N- X2+1 new reads for each read of sequence length N, wherein the new N- X2+1 reads correspond to all k-mers with length X2 nucleotides; and

C2) Building a hashtable structure, which comprises all the k-mers generated in step C1 ) and further comprises the number of times each k-mer is observed in the two sets of reads corresponding to first and second states.

D) Computer/Electronic means for detecting variants in the sequence between first state and second state, wherein a k-mer of the hashtable structure is taken as a candidate breakpoint, which represents a variant between the first and second states, if it fulfills all the following requirements: D1 ) At least one inflection based on a k-mer's stem must have at least X3 reads with the same variation between first and second states, being X3 at least 2; D2) The percentage of first state reads in second state reads is not over a threshold X4, to account for possible contamination of control state reads with test state reads, wherein X4 is at least 5%. E) Computer/Electronic means for clustering and filtering test and control reads derived from all candidate breakpoints accepted in step D to build blocks, by carrying out the steps: E1 ) Retrieving reads which contain the stem of at least one k-mer that represents the candidate breakpoint selected in step D); E2) The reads of step E1 ) with at least X5 k-mer variants within a window of X6 nucleotides are taken as leading reads, wherein X5 is at least 7 and X6 is at least 10; E3) Reads whose k-mers share at least one stem with a leading read are merged to give a block; and E4) If the nucleic acid whose variant is being identified is a double stranded DNA, then both forward and reverse variants are taken into account when building the block. F) Computer/Electronic means for aligning blocks taking their leading reads as a reference: F1 ) For each read in the block, take the leading read's stem and find the longest inflection or partial inflection between the read and the leading read. F2) successively position each read so that its matching inflection or partial inflection is aligned against the leading read. The electronic/computer means may be used interchangeably, that is, a part of the described means may be electronic means and the other part may be computer means, or all described means may be electronic means or all described means may be computer means. Examples of an apparatus comprising only electronic means may be a CPLD (Complex Programmable Logic Device), a FPGA (Field Programmable Gate Array) or an ASIC

(Application-Specific Integrated Circuit). A fourth aspect of the invention is a computer system comprising a processor and a memory, wherein the memory stores computer executable instructions that, when executed by the processor, cause the system to perform the method for identificatying nucleic acid variants between two genomic states. In some examples, the computer system may further comprise a hardware accelerator.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 . Data flow for the proposed configuration of the system. 1 .1 Quality Check; 1 .2 Base Convertion; 1 .3 Reduce; 1 .4 Hash; 1 .5 Orchestable Data Movement; 1 .6 Store Data into Associative Structure; 1 .7 Load Sequences and Quality Markers; 1 .8 Network; 1 .9 Disk(s).

FIG2. Filtering capabilities of the method at different stages, -o- is Total Reads; -x-is Supporting Reads; -I- is Identifiable Mutations.

FIG3. Visualization of a breaking-block with an insertion of virus

DETAILED DESCRIPTION OF THE INVENTION

All terms as used herein, unless otherwise stated, shall be understood in their ordinary meaning as known in the art. Other more specific definitions for certain terms as used in the present application are as set forth below and are intended to apply uniformly throughout the description and claims unless an otherwise expressly set out definition provides a broader definition.

It must be noted that for clarity reasons, a variety of nucleotide sequences are given in the definitions and the examples found below. These nucleotide sequences are made up for these examples and they do not refer to any real nucleotide sequence of any organism. They are only listed so that the reader of the present application understands the terms used such as k-mer, stem, inflection, partial inflection, etc. They are absolutely unrelated to the invention being disclosed herewith, which has nothing to do with any particular nucleotide sequence. The terms "computational method" and "computer implemented method" are taken here to mean the same and are used interchangeably. Therefore, "computational method" and "computer implemented method" are taken as synonyms.

The terms "Next Generation Sequencing" (NGS), "deep sequencing", "ultra- deep sequencing", "high throughput sequencing" are all used interchangeably and refer to the technology platforms currently being used as standard to enable the sequencing of genomes ("sequencing methods") with high speed and contained cost, such as the Roche/454, lllumina/Solexa, Life/APG and Pacific Biosciences platforms.

The term "base" and the term "nucleotide" are herein used interchangeably, and refer to the monomers (subunits) which are repeated in a nucleic acid such as DNA or RNA, giving its sequence or primary structure.

The term "reference genome" as used herein refers to the complete nucleic acid sequence representing the whole genome of a species normally accepted by the wide community. Since the reference genome is usually assembled from the sequencing of DNA from a number of donors, it does not accurately represent the set of genes of any one single individual. Instead, a reference genome provides a mosaic of different DNA sequences from each donor. But, at general levels, the reference genome provides a good approximation of the DNA of any single individual. However, in genomic regions with high allelic diversity, the reference genome may differ

significantly from any one single individual. For example, GRCh37, the Genome Reference Consortium human genome (build 37) is derived from thirteen anonymous volunteers from New York. Reference genomes are typically used as a guide on which new genomes are built and aligned, enabling their assembly and comparison.

The term "forward strand" as used herein refers to a nucleic acid sequence read from 5' terminal to 3' terminal ends. The term "reverse strand" refers to the nucleic acid sequence which is complementary to the forward strand. The term "nucleic acid variant" or simply "variant" as used herein refers to a difference in sequence between two genomic states. A variant can be a single nucleotide variant (SNV) if the difference between the two genomes (or two states of the same genome) is only due to the change of a single nucleotide. All other variants, among them insertions, deletions, inversions, duplications, translocations and others are termed structural variants (SV). The latter can have many sizes, from two bases up to entire pieces of a chromosome.

The term "genomic state" as used herein can refer to two different genomes derived from two different individuals, or two genomes derived from two different cells of the same individual. In the second case, the two different cells can be a normal vs. a pathological cell, an undifferentiated vs. a differentiated cell, a cell which has been exposed to a certain external factor vs. an unexposed cell, etc.

The term "mapping" as used herein refers to aligning blocks of the first and second state to a reference genome.

The term "read" as used herein refers to a fragment of nucleic acid that is sequenced in its entirety. The nucleic acid might be DNA, RNA, or even chemically altered nucleic acids. The initial step in a high throughput sequencing run is the random fragmentation of a genome into millions of partly overlapping fragments called reads, which are usually amplified by Polymerase Chain Reaction and sequenced using a variety of techniques that are platform-dependent. The lengths of the reads can also vary depending on the platform, and are usually on the order of a few dozens to a few hundreds of nucleotides. The partly overlapping reads must be assembled if a complete picture of the genome is to be built. The term "depth of coverage" as used herein refers to the number of times a nucleotide is read during the sequencing process. Deep sequencing means that the total number of reads is many times larger than the length of the sequence under study. Standard depth of coverage currently range from 30x to 10Ox for whole genomes, meaning that each position in the genome is represented from 30 to 100 times. Coverage similarly designates the average number of reads representing a given nucleotide in the reconstructed sequence.

Depth of coverage can be calculated from the length of the original genome (G), the number of reads (A/), and the average read length (L) as N ^* L/G.

The term "undefined nucleotide" as used herein refers to a certain position inside a sequenced read that could not be determined during the sequencing process, that is, a position for which the sequencing experiment has not unambiguously resolved whether it is occupied by an adenine (A), guanine (G), cytosine (C) or thymine (T), and therefore its nature is unknown.

Undefined nucleotides in reads are filtered out (removed) in the method of the invention, generating two or more fragments of defined sequence if the undefined nucleotides are removed form inner positions of the read. The term "Phred quality score" as used herein refers to the quality score given to each nucleotide base call in a sequenced read. The Phred score is a property given to each sequenced nucleotide and it is logarithmically related to the base-calling error probability. A Phred score of 10 assigned to a certain nucleotide in a sequenced read means that there is a 90% probability that the base call is correct, a Phred score of 20 means that there is a 99% probability that the base call is correct, and a Phred score of 30 means that there is a 99.9% probability that the base call is correct.

The term "assembling" as used herein refers to grouping all the first state reads, and separately second state reads that share the same variant.

The term "hashtable" as used herein refers to a data structure used in computing that allows (by applying a hash function) assigning and mapping hashes (values) to strings of data, that is, it associates a series of hashes (values) to a series of strings in pairs, such that the association of hash-string is established. The addition, removal and modification of pairs is easily achieved by computational means, as well as the lookup and accession of strings thanks to their respective hashes (values). The term "hash" as used herein refers to the value given by the hash function and linked to a certain string. This value allows computationally storing, retrieving, deleting and sorting strings in a very efficient manner.

The term "hash function" as used herein refers to the function that is used for linking hashes (values given by the hash function) to strings of data given as input.

The term "k-mer" as used herein refers to all possible substrings of length k that are contained in a string. In genomics, all k-mers of a nucleic acid read are all the possible sub-sequences within the original read which have a length k. The amount of k-mers in a read of length M is M-k+1 .

The term "polymorphic k-mer" as used herein refers to a k-mer that also identifies inflections and partial inflections of the k-mer's stem. By

polymorphic k-mer it is here understood the way the method of the invention handles k-mers, that is, the way they are used to derive stems and the latter to search for, manipulate and fish inflections and partial inflections.

The term "stem" as used herein refers to a fragment of a k-mer of length k with S defined bases, where S < k, and k-S omitted (undefined) bases. The stem fragment can either be a k-mer without a prefix, a k-mer without a suffix, a k-mer without an infix, or any combination thereof. Stem fragments without infix and/or prefix are consecutive, while stems without infixes can be non- consecutive. In a stem, the character "-" denotes a base that is omitted from the k-mer. Examples of the 30-mer SEQ ID NO: 1

CACGGCAGCTGAGTCAACAGGTTCTTCCCA:

SEQ ID NO:2

CACGGCAGCTGAGTCAACAGGTTCTTCCC- (omission of suffix of length 1 ) SEQ ID NO:3

-ACGGCAGCTGAGTCAACAGGTTCTTCCC- (omission of prefix of length 1 , and suffix of length 1 )

SEQ ID NO:4

CACG--AGCTGAGTCAACAGGTTCTTCCCA (omission of infix of length 2 starting at position 5) The term "prefix" as used herein, refers to the first part of a sequencing read, that is, from position 1 to a given position depending on the context. This term is used here, as it is used in a grammatical context referring to words.

The term "suffix" as used herein, refers to the last part of a sequencing read, that is, from the last position to a given position depending on the context. This term is used here, as it is used in a grammatical context referring to words.

The term "infix" as used herein, refers to a part of a read positioned in the middle of the sequence.

The term "inflection" as used herein refers to a fragment of length k that can be derived from extending a stem of length k-1 , k-2, k-3, etc. of a k-mer of length k. E.g. A stem of length k-1 can be used to derive 4 inflections of length k since there is a single unknown position, and 4 different bases (4 ¹ =4). A stem of length k-2 can be used to derive 16 inflections of length k (4 ² =16). Following the example given above, for the stem (SEQ ID NO:5): "CACGGCAGCTGAGTCAACAGGTTCTTCCC-" (omission of suffix of length 1 ) the inflections would be (SEQ ID NO:6 to SEQ ID NO:9):

CACGGCAGCTGAGTCAACAGGTTCTTCCCA CACGGCAGCTGAGTCAACAGGTTCTTCCCC CACGGCAGCTGAGTCAACAGGTTCTTCCCT CACGGCAGCTGAGTCAACAGGTTCTTCCCG

and further,

inflections based on stem (SEQ ID NO:10)

"-ACGGCAGCTGAGTCAACAGGTTCTTCCC-" would be (SEQ ID NO:1 1 -

SEQ ID NO:26):

AACGGCAGCTGAGTCAACAGGTTCTTCCCA AACGGCAGCTGAGTCAACAGGTTCTTCCCC AACGGCAGCTGAGTCAACAGGTTCTTCCCT AACGGCAGCTGAGTCAACAGGTTCTTCCCG

CACGGCAGCTGAGTCAACAGGTTCTTCCCA CACGGCAGCTGAGTCAACAGGTTCTTCCCC CACGGCAGCTGAGTCAACAGGTTCTTCCCT CACGGCAGCTGAGTCAACAGGTTCTTCCCG TACGGCAGCTGAGTCAACAGGTTCTTCCCA TACGGCAGCTGAGTCAACAGGTTCTTCCCC TACGGCAGCTGAGTCAACAGGTTCTTCCCT

TACGGCAGCTGAGTCAACAGGTTCTTCCCG GACGGCAGCTGAGTCAACAGGTTCTTCCCA GACGGCAGCTGAGTCAACAGGTTCTTCCCC GACGGCAGCTGAGTCAACAGGTTCTTCCCT GACGGCAGCTGAGTCAACAGGTTCTTCCCG

The term "partial inflection" as used herein refers to a fragment with P defined bases that can be derived from extending a stem of S defined bases of a k- mer of length k, and where S<=P<k. In a partial inflection, the «.» character denotes a non-extended position of its stem. Partial inflections must have at least one non-extended position. Only omitted bases («-») can be marked as non-extended.

Following the example given above:

Partial inflections based on stem (SEQ ID NO:27)

"-ACGGCAGCTGAGTCAACAGGTTCTTCCC-" would be (SEQ ID NO:28 - SEQ ID NO:36):

.ACGGCAGCTGAGTCAACAGGTTCTTCCC.

AACGGCAGCTGAGTCAACAGGTTCTTCCC.

CACGGCAGCTGAGTCAACAGGTTCTTCCC.

TACGGCAGCTGAGTCAACAGGTTCTTCCC.

GACGGCAGCTGAGTCAACAGGTTCTTCCC.

.ACGGCAGCTGAGTCAACAGGTTCTTCCCA

.ACGGCAGCTGAGTCAACAGGTTCTTCCCC

.ACGGCAGCTGAGTCAACAGGTTCTTCCCT

.ACGGCAGCTGAGTCAACAGGTTCTTCCCG

The term "breakpoint" as used herein refers to the the nucleotide position where the sequence changes, that is sequence immediately flanking a sequence variant. For SVs, a breakpoint is the point where the DNA broke in the second state and appears as a change in the first state compared to the second control state. In other words, where the continuity of the sequence of the control second state breaks (changes) in the first state. The term "leading read" as used herein refers to a complete sequenced read that contains at least one k-mer that is a candidate breakpoint (variant). Normally, in the case of heterozygous variation, only the reads derived from the altered allele contain the mutation or variant.

The term "block" refers to a leading read along with all reads derived from the sequencing of all four alleles involved (two coming from the first state genome and two coming from the second state genome) covering the same region as the leading read.

The term "similarity score" as used herein refers to numbers that help to identify how different sets of aligned sequences are, and can be used as part of the proposed method to measure the quality of an aligned block. Similarity scores can be vertical or horizontal. The former measures, for every position in a sequence, how many bases in the set of aligned sequences are different than the mode base. The latter measures, for every sequence in the set, how many positions of the sequence are different to the mode/consensus sequence. Similarity scores can be measured for different sets of sequences, e.g. the set of control sequences, the set of test sequences, or the set containing both. The term "ambiguous path" as used herein refers to multiple possible sequence solutions in a given tree. It is referred here as the opposite of unique and unambiguous path or sequence.

The terms defined above are used in the following example for increasing their clarity and conciseness:

Imagine the read to be input:

A) (SEQ ID NO: 37)

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGGTGGC CAGCAGGCACGTG (its length N=61 ).

After the quality filtering (step B) of the method), the next step of the method C) would be to generate a hashtable with all the k-mers of length X2. If X2 is taken to be 30, then, there should be N-X2+1 30-mers, that is, 61 -30+1 =32 (SEQ ID NO:38 - SEQ ID NO:69):

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGGT GGCCAGCAGGCACGTG

CACGGCAGCTGAGTCAACAGGTTCTTCCCA ACGGCAGCTGAGTCAACAGGTTCTTCCCAG CGGCAGCTGAGTCAACAGGTTCTTCCCAGG GGCAGCTGAGTCAACAGGTTCTTCCCAGGA GCAGCTGAGTCAACAGGTTCTTCCCAGGAG CAGCTGAGTCAACAGGTTCTTCCCAGGAGC AGCTGAGTCAACAGGTTCTTCCCAGGAGCG GCTGAGTCAACAGGTTCTTCCCAGGAGCGG CTGAGTCAACAGGTTCTTCCCAGGAGCGGA TGAGTCAACAGGTTCTTCCCAGGAGCGGAC GAGTCAACAGGTTCTTCCCAGGAGCGGACG AGTCAACAG GTTCTTCCCAG G AGCG G ACG G GTCAACAGGTTCTTCCCAGGAGCGGACGGC TCAACAGGTTCTTCCCAGGAGCGGACGGCG CAACAGGTTCTTCCCAGGAGCGGACGGCGG AACAGGTTCTTCCCAGGAGCGGACGGCGGT ACAGGTTCTTCCCAGGAGCGGACGGCGGTG CAGGTTCTTCCCAGGAGCGGACGGCGGTGG AGGTTCTTCCCAGGAGCGGACGGCGGTGGC GGTTCTTCCCAGGAGCGGACGGCGGTGGCC GTTCTTCCCAGGAGCGGACGGCGGTGGCCA TTCTTCCCAGGAGCGGACGGCGGTGGCCAG TCTTCCCAGGAGCGGACGGCGGTGGCCAGC CTTCCCAGGAGCGGACGGCGGTGGCCAGCA TTCCCAGGAGCGGACGGCGGTGGCCAGCAG TCCCAGGAGCGGACGGCGGTGGCCAGCAGG CCCAGGAGCGGACGGCGGTGGCCAGCAGGC CCAGGAGCGGACGGCGGTGGCCAGCAGGCA CAGGAGCGGACGGCGGTGGCCAGCAGGCAC AGGAGCGGACGGCGGTGGCCAGCAGGCACG GGAGCGGACGGCGGTGGCCAGCAGGCACGT GAGCGGACGGCGGTGGCCAGCAGGCACGTG

The hashtable to be generated with all k-mers and their number of times they are observed in first state and second state, would look like (SEQ ID NO:70 - SEQ ID NO:75):

D) The next step in the method would be to detect variants between the first and second states. The first step would be to derive a stem for each one of the k-mers (SEQ ID NO:76 - SEQ ID NO:108):

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGGT GGCCAGCAGGCACGTG

CACGGCAGCTGAGTCAACAGGTTCTTCCC- ACGGCAGCTGAGTCAACAGGTTCTTCCCA- CGGCAGCTGAGTCAACAGGTTCTTCCCAG- GGCAGCTGAGTCAACAGGTTCTTCCCAGG- GCAGCTGAGTCAACAGGTTCTTCCCAGGA- CAGCTGAGTCAACAGGTTCTTCCCAGGAG- AGCTGAGTCAACAGGTTCTTCCCAGGAGC- GCTGAGTCAACAGGTTCTTCCCAGGAGCG- CTGAGTCAACAGGTTCTTCCCAGGAGCGG- TGAGTCAACAGGTTCTTCCCAGGAGCGGA- GAGTCAACAGGTTCTTCCCAGGAGCGGAC- AGTCAACAGGTTCTTCCCAGGAGCGGACG- GTCAACAGGTTCTTCCCAGGAGCGGACGG- TCAACAGGTTCTTCCCAGGAGCGGACGGC- CAACAGGTTCTTCCCAGGAGCGGACGGCG- AACAGGTTCTTCCCAGGAGCGGACGGCGG- ACAGGTTCTTCCCAGGAGCGGACGGCGGT- CAGGTTCTTCCCAGGAGCGGACGGCGGTG- AGGTTCTTCCCAGGAGCGGACGGCGGTGG- GGTTCTTCCCAGGAGCGGACGGCGGTGGC- GTTCTTCCCAGGAGCGGACGGCGGTGGCC- TTCTTCCCAGGAGCGGACGGCGGTGGCCA- TCTTCCCAGGAGCGGACGGCGGTGGCCAG- CTTCCCAGGAGCGGACGGCGGTGGCCAGC- TTCCCAGGAGCGGACGGCGGTGGCCAGCA- TCCCAGGAGCGGACGGCGGTGGCCAGCAG- CCCAGGAGCGGACGGCGGTGGCCAGCAGG- CCAGGAGCGGACGGCGGTGGCCAGCAGGC- CAGGAGCGGACGGCGGTGGCCAGCAGGCA- AGGAGCGGACGGCGGTGGCCAGCAGGCAC- GGAGCGGACGGCGGTGGCCAGCAGGCACG- GAGCGGACGGCGGTGGCCAGCAGGCACGT-

For each stem, inflections are to be generated. For instance, the second stem found in the table above would give the following 4 inflections (SEQ ID NO:109 - SEQ ID NO:1 13):

Find each one of the inflections in the hashtable:

K-mer Normal Tumor

ACGGCAGCTGAGTCAACAGGTTCTTCCCAA 0 1

ACGGCAGCTGAGTCAACAGGTTCTTCCCAC 0 4

ACGGCAGCTGAGTCAACAGGTTCTTCCCAT 0 1

ACGGCAGCTGAGTCAACAGGTTCTTCCCAG 0 1 If one inflection meets the requirements (has at least X3 reads with the same variation between the first and second states, and the amount of first state reads in second state reads is not over X4), select k-mer as candidate breakpoint:

E) Clustering and filtering test and control reads derived from all candidate breakpoints:

For each candidate breakpoint selected in step D), retrieve reads which contain the stem of the k-mer (SEQ ID NO:1 14 - SEQ ID NO:1 16):

Reads for candidate breakpoint based on k-mer

AGTCAACAGGTTCTTCCCAGGAGCGGACGC

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGG TGGCCAGCAGGCACGTG

AGCAGGCACGTGACGGCAGCTGCAGTCAACAGGTTCTTCCCAGG AGCGGACGCCGGTGGCC

Then, for each read, we will end up with all k-mers that are candidate breakpoints, and their positions in the read (SEQ ID NO:1 17 - SEQ ID NO:126):

Read #1

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGG TGGCCAGCAGGCACGTG

ACGGCAGCTGAGTCAACAGGTTCTTCCCAG (1 ) GGCAGCTGAGTCAACAGGTTCTTCCCAGGA (3)

GCAGCTGAGTCAACAGGTTCTTCCCAGGAG (4)

CAGCTGAGTCAACAGGTTCTTCCCAGGAGC (5)

AGCTGAGTCAACAGGTTCTTCCCAGGAGCG (6)

GCTGAGTCAACAGGTTCTTCCCAGGAGCGG (7)

TGAGTCAACAGGTTCTTCCCAGGAGCGGAC (9) GAGTCAACAGGTTCTTCCCAGGAGCGGACG (10) AGTCAACAGGTTCTTCCCAGGAGCGGACGG (1 1 )

(SEQ ID NO:127 - SEQ ID NO:128)

Read #2

AGCAGGCACGTGACGGCAGCTGCAGTCAACAGGTTCTTCCCAGG AGCGGACGCCGGTGGCC

AGTCAACAG GTTCTTCCCAG G AGCG G AC GC (23)

Reads with (X5) 7 k-mers containing candidate breakpoints within a window of (X6) 10 nucleotides are taken as leading reads (SEQ ID NO:129).

Leading reads

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGG TGGCCAGCAGGCACGTG

Generate stems from leading reads (SEQ ID NO130 - SEQ ID NO:138):

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGG TGGCCAGCAGGCACGTG

ACGGCAGCTGAGTCAACAGGTTCTTCCCA- GGCAGCTGAGTCAACAGGTTCTTCCCAGG- GCAGCTGAGTCAACAGGTTCTTCCCAGGA- CAGCTGAGTCAACAGGTTCTTCCCAGGAG- AGCTGAGTCAACAGGTTCTTCCCAGGAGC- GCTGAGTCAACAGGTTCTTCCCAGGAGCG- TGAGTCAACAGGTTCTTCCCAGGAGCGGA- GAGTCAACAGGTTCTTCCCAGGAGCGGAC- AGTCAACAGGTTCTTCCCAGGAGCGGACG-

Build blocks by finding other candidate reads that share stems with leading reads (SEQ ID NO: 139 - SEQ ID NO:141 ):

Leading read #1

CACGGCAGCTGAGTCAACAGGTTCTTCCCAGGAGCGGACGGCGG TGGCCAGCAGGCACGTG

Additional reads

AGCAGGCACGTGACGGCAGCTGCAGTCAACAGGTTCTTCCCAGG AGCGGACGCCGGTGGCC

Shared stems

AGTCAACAGGTTCTTCCCAGGAGCGGACG-

By executing the steps outlined above in the right order, the application of the computer-implemented method of the invention allows to easily cut the huge numbers of reads given in the input down to a much reduced number where most of the true positives are found, as will be seen in the experimental data found in the examples section (below).

As is revealed in Paszkiewicz K., et.al. "De novo assembly of short sequence reads" Brief Bioinform. 2010, vol. 1 1 , pp. 475-472, the approximate minimum length that NGS reads must have in order to be able to reconstruct a genome is around 30. Bearing in mind the latter, a minimum length of approximately 30 bases was taken to be the minimum length of a productive read. 30 is a value that was found to be a viable cutoff for variable X2 in the definition of the method of the invention, although slightly smaller values might be viable as well.

As it has been cited above, the first aspect of the present invention is a computer-implemented method for identifying nucleic acid variants between two genomic states comprising the steps of: A) Inputting 2 sets of nucleic acid reads, which are sequences retrieved from a nucleotide sequencing method, wherein the first set of reads corresponds to cells representing a first test state, and the second set of reads corresponds to cells representing a second control state;B) Filtering the reads, wherein the filtering comprises: B1 ) Keeping only the reads with at least a percentage X1 of their bases with a Phred quality score higher than 20, being X1 equal to or above 90%; B2) Splitting the reads with an undefined nucleotide, giving one sequence before, and one sequence after the undefined nucleotide, the latter being discarded; and B3) Discarding the sequence reads with less than X2 bases, wherein X2 is from 25 to 50; C) Generating a hashtable structure comprising: C1 )

Generating a number of N-X2+1 new reads for each read of sequence length N, wherein the new N-X2+1 reads correspond to all k-mers with length X2 nucleotides; and C2) Building a hashtable structure, which comprises all the k-mers generated in step C1 ) and further comprises the number of times each k-mer is observed in the two sets of reads corresponding to first and second states. D) Detecting candidate variants in the sequence between first state and second state, wherein a k-mer of the hashtable structure is taken as a candidate breakpoint, which represents a variant between the first and second states, if it fulfills all the following requirements: D1 ) At least one inflection based on a k-mer's stem must have at least X3 reads with the same variation between first and second states, being X3 at least 2; 2) The percentage of first state reads in second state reads is not over a threshold X4, to account for possible contamination of control state reads with test state reads, wherein X4 is at least 5%. E) Clustering and filtering test and control reads derived from all candidate breakpoints accepted in step D to build blocks, by carrying out the steps: E1 ) Retrieving reads which contain the stem of at least one k-mer that represents the candidate breakpoint selected in step D); E2) The reads of step E1 ) with at least X5 k-mer variants within a window of X6 nucleotides are taken as leading reads, wherein X5 is at least 7 and X6 is at least 10; E3) Reads whose k-mers share at least one stem with a leading read are merged to give a block; and E4) If the nucleic acid whose variant is being identified is a double stranded DNA, then both forward and reverse variants are taken into account when building the block. F) Aligning blocks taking their leading reads as a reference: F1 ) For each read in the block, take the leading read's stem and find the longest inflection or partial inflection between the read and the leading read. F2) Successively position each read so that its matching inflection or partial inflection is aligned against the leading read.

In a particular embodiment of the first aspect of the invention, the computer- implemented method further comprises the step following F2: F3) F3)

Obtaining first state scores, second state scores, and global similarity scores of each position in the block by measuring a ratio of most frequent nucleotide in that position relative to the total number of nucleotides. In a particular embodiment of the first aspect of the invention, the computer- implemented method further comprises the step: G) Cataloguing and annotating blocks according to the following: G1 ) If blocks between the first and second states only differ in one substituted nucleotide, the variant is catalogued as containing a single nucleotide variant and the single nucleotide variant is annotated; G2) If blocks between the first and second states differ in more than one nucleotide but the whole difference in sequence is contained within the block, the variant is catalogued as a small structural variant, and the small structural variant is annotated; and G3) If blocks between the first and second states differ in more than one nucleotide and the whole difference in sequence is not contained within the block, the variant is catalogued as a large structural variant, and the boundaries of all large structural variants are extended by retrieving blocks overlapping at least X2 nucleotides in an iterative process which ends when the extended sequence reaches 200 nucleotides or when an ambiguous path is found.

In a particular embodiment of the first aspect of the invention, the computer- implemented method further comprises the step: H) Filtering of the blocks, according to the following: H1 ) The percentage of second state reads in first state reads is not over a threshold X7, to account for possible contamination of test state reads with control state reads, wherein X7 is at least 20%;

In a particular embodiment of the first aspect of the invention, the method further comprises optionally mapping second state blocks, and subsequently mapping first state blocks, on a reference genome.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X1 is equal or above 95%.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X1 is equal or above 99%.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X2 is from 25 to 40.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X2 is from 30 to 35.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X2 is from 30 to 32. In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X2 is equal to 30.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X3 is equal or above 4.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X3 is equal or above 6.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X3 is equal or above 8.

In a particular embodiment of the first aspect of the invention, X3 is directly proportional to the depth of coverage in the sequencing experiment. This means that, the deeper the coverage, the more restrictive (higher) is the value for X3.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X4 is between 5-10%. In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X4 is between 5-7%. In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X4 is 5%. X4 is expressed as a percentage, reflecting the maximum accepted ratio of first state (test) reads vs. second state (control) reads for each of the k-mers, and represents the levels of contamination expected for each of the samples (usually in the direction of tumor cells within normal samples). This value should be set by the user accordingly. Setting up a low value for X4 ensures high specificity but might result in a lower sensitivity, whereas the selection of high values, might result in the accumulation of false positives.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X5 is from 10 to 15.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X5 is from 12 to 14.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X6 is from 12 to 25.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X6 is from 12 to 20. In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X6 is from 12 to 15.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X7 is between 20-25%.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X7 is 20%.

In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X5 is from 10 to 15 and the threshold X6 is from 12 to 20. In a particular embodiment of the first aspect of the invention, optionally in combination with any embodiment above or below, X5 is from 12 to 14 and the threshold X6 is from 12 to 15.

In a particular embodiment of the first aspect of the invention, the first set of reads corresponds to pathological cells of a patient, and the second set of reads corresponds to non-pathological cells of the same patient; In another particular embodiment of the first aspect of the invention, the first set of reads corresponds to cancer cells of a patient, and the second set of reads corresponds to non-cancer cells of the same patient.

In another particular embodiment of the first aspect of the invention, the first set of reads corresponds to virus-infected cells of a patient, and the second set of reads corresponds to non-infected cells of the same patient.

In another particular embodiment of the first aspect of the invention, the first set of reads and the second set of reads correspond to the same cell of the same patient in two different developmental stages.

In another particular embodiment of the first aspect of the invention, the first set of reads corresponds to cells of a patient which have been exposed to a drug, and the second set of reads corresponds to cells of the same patient which have not been exposed to a drug.

In a particular embodiment of the first aspect of the invention, the first set of reads corresponds to cells of a tissue, and the second set of reads corresponds to cells of another tissue of the same or a different individual.

Although the present invention has been described in detail for purpose of illustration, it is understood that such detail is solely for that purpose, and variations can be made therein by those skilled in the art without departing from the scope of the invention.

Thus, while the preferred embodiments of the methods and of the systems have been described in reference to the environment in which they were developed, they are merely illustrative of the principles of the invention. Other embodiments and configurations may be devised without departing from the scope of the appended claims.

Further, although the embodiments of the invention comprise processes performed in computer systems, the invention also extends to computer systems and to computer programs (which may be embodied on a storage medium and/or carried on a carrier signal) adapted for putting the invention into practice.

Accordingly, the invention also provides a computer program product comprising program instructions for causing a computer system to perform the method for identifying nucleic acid variants between two genomic states as defined above.

In a preferred embodiment, the computer program product is embodied on a storage medium. In another preferred embodiment, the computer program product is carried on a carrier signal. The carrier may be any entity or device capable of carrying the program.

As it has been cited above, a fourth aspect of the invention is a computer system comprising a processor and a memory, wherein the memory stores computer executable instructions that, when executed by the processor, cause the system to perform the method for identifying nucleic acid variants between two genomic states. In a preferred embodiment of the fourth aspect of the invention, the system may further comprise a hardware accelerator, which may be in some examples an FPGA or a GPU.

Throughout the description and claims the word "comprise" and variations of the word, are not intended to exclude other technical features, additives, components, or steps. Furthermore, the word "comprise" and its variations encompasses the term "consisting of. Additional objects, advantages and features of the invention will become apparent to those skilled in the art upon examination of the description or may be learned by practice of the invention. The following examples are provided by way of illustration, and they are not intended to be limiting of the present invention. Furthermore, the present invention covers all possible combinations of particular and preferred embodiments described herein.

EXAMPLES

Examples of using the method of the invention for detecting characterizing sequence variants in nucleic acid sequences are given below.

In the in silico tests it is revealed that the method of the invention is capable of identifying SNVS and SVs of all sorts and sizes. Remarkably, the method of the invention is even proven to be capable of identifying novel non-human insertions. In one of the examples found below, the method is remarkably capable of detecting the insertion of a virus where, other methods (including the one disclosed in Moncunill et al., ibid) fail.

Material and methods

An implementation of the computer-implemented method The general structure and the complete variant identification and

characterization carried out by the method of the invention comprise the steps outlined below:

A) Input data.

As input, the method takes high quality sequences data directly from FASTQ files of tumor and non-tumor control cells samples of the same individual. Alternatively, it is also able to accept BAM files, from which it extracts all the sequencing reads. Tumor sample corresponds to the first state and non-tumor control sample correspond to the second state.

B) Filtering the data. When inputting the data, the user can define a cut-off so that reads having over a certain threshold of their bases with a Phred quality score <q20 are discarded. X1 =90 has been found to be especially suited for the purposes tested. This means that only reads with at least 90% of their bases with a Phred quality score higher than 20 are kept. In the case of the presence of undefined base pairs ("N"), these are removed and the original sequence is split forming new shorter reads, which are considered only if they are longer than X2 base pairs. In order to lower the amount of space needed to store k-mers, they are converted into integers by mapping each base of the k-mer to a 2-bit code. For instance, a k-mer of length 32 represented as a sequence of characters takes 256 bits, but after the conversion it is turned into a single value of 64 bits.

In a particular embodiment of the computer-implemented method, wherein hardware accelerator(s) may be used,, reads that don't meet the

aforementioned quality criteria are discarded by means of marking their k- mers as discardable, which are then ignored in the subsequent steps of the pipeline. Marking k-mers as discardable instead of deleting them immediately enables a faster pipeline without conditional execution.

C) Generating a hashtable structure. After the quality filtering step of the method, the next step of the method is to generate a hashtable with all the k-mers of length X2 using all high quality tumor and non-tumor control reads (see Table 1 below for a simplified version). If X2 is taken to be 28, then, there should be N-X2+1 , that is, 100- 28+1 =73 resultant k-mers for a 100-nucleotide read. Each of k-mers generated is inserted into the hashtable and their number of times they are observed in tumor (test state) and non-tumor (second control state) cells.

The mapping of k-mers to their observed frequencies in the input is, generally speaking, one to one, meaning each entry of the associative data structure contains a pair of frequencies. In order to find the position where to store data into an associative structure a hash function is used to compute an index into a position, from which the desired value can be stored and retrieved. Any function that guarantees a homogeneous distribution of the results can be used as hash function.

Table 1

(SEQ ID NO:142 - SEQ ID NO:145)

Each item of the hashtable consist of its k-mer (in nucleotide string or encoded key format), along with its pair frequencies in the first state and second state sets of reads (Table 1 ).

In particular embodiments of this aspect, each entry of the associative data structure may contain frequencies for more than one k-mer. This is accomplished by means of indexing stems instead of k-mers. Table 2 below stem (length Χ2-1Ϊ Frequencies

ACTGACTGACTGACTGACTGACT6ACTGA Α : ( θ , 1 ) C (i,&) G: ( 23 , 2 ) ?: (§ , !§} ACTGACTGACT6ACTGACTGACTGACTGC

depicts such an example, where stems are basically the original k-mer truncating the last base, which is then included as part of the list of frequencies. Indexing stems instead of k-mers improves the locality of the data, reducing the number of lookup queries. Table 2 (SEQ ID NO:146 and SEQ ID NO:147)

In a particular embodiment, the hash function operates over the encoded key as described in step B instead of the k-mer containing a string of nucleotides. In order to lower the number of updates to the associative data structure, a particular embodiment of this aspect involves generating an additional structure to store the set of k-mers seen only once, meaning the main data structure is only updated for k-mers with a frequency of 2 or more. In another embodiment, a lower number of updates to the main data structure is achieved by generating partial data structures containing frequencies for a subset of the input, which are then merged into the main associative data structure. D) Detecting k-mers containing variants once all the reads are derived on k- mers and loaded into the hashtable structure, the next step consists in identifying all tumor specific reads. Inventors expect that variants generate new and distinct sequences in the tumor genome compared to the non- mutated control genome.

If one inflection meets the requirements: has at least 4 (X3) k-mers with the same variation between tumor cells and maximum 1 (X4) k-mer non-tumor control cells, then this k-mer is select as a candidate breakpoint. The detection of entries of the hashtable containing k-mers with variants between the first and second state is accomplished by reading the filtered input again, and selecting all reads that contain k-mers whose frequencies meet certain criteria. In addition to selecting candidate variants, in this step the implementation also selects additional information needed during later steps of the pipeline, namely: 1 ) selection of relative reads; 2) position of candidate k-mers for each read; and 3) map of k-mers to reads. In particular, the selection of relative reads is done based on stems, and checking whether any inflection matches the criteria described in the previous paragraph. E) Clustering and filtering test and control reads

For each candidate breakpoint selected in step D), retrieve reads which contain the stem of the k-mer. Then for each read, we will end up with all k- mers that are candidate breakpoints, and their positions in the read. Reads with 7 (X5) k-mers containing candidate breakpoints within a window of 10 (X6) nucleotides are taken as leading reads. This process is in order to find enough support information for each breakpoint detected (if not it is removed from the candidate list). Reads whose k-mers share at least one stem with a leading read are merged to give a block, and if the nucleic acid whose variant is being identified is a double stranded DNA, then both forward and reverse variants are taken into account when building the block.

F) Aligning blocks taking their leading reads as a reference.

For each read in the block, inventors take the leading read's stem and find the longest inflection or partial inflection between the read and the leading read. Successively position each read so that its matching inflection or partial inflection is aligned against the leading read.

Ideally each block represents a region in the genome containing the mutated and the non-mutated version. In order to classify and characterize the type of variation identified, the method takes into account the align score for tumor block, non-tumor block and the sum up of both. Then the method extracts the consensus mutated and normal sequences from these blocks. The

corresponding normal consensus sequence can be used at the end of the procedure and mapped onto a reference genome to obtain the coordinates of the variant.

Optionally the method also can include step G

G) Cataloguing and annotating blocks

Once all possible breakpoint blocks are defined, the next step consists in identifying and classifying the variation included there. At this point the method uses the aligning score to observe the differences in each group: tumor, non-tumor and both. For each position on the block supported by a read it puts a value depending on the similarity, so finally it has a

representation of all the variability on the block. These aligning scores are recursively compared to identify differences between tumor and non-tumor samples. A first evaluation will search a consensus in non-tumor block in order to avoid false positives and wrong alignments from different regions, and the same way on tumor blocks. The next step searches for all the small variants, which consist on those that are completely included within the block (SNV and small SVs: insertions, deletions and inversions). All the blocks that do not match this criterion are then considered candidates for large SVs, i.e. likely to cover break points of intra or interchromosomal transitions, part of large deletions, insertions or inversions. After small and large somatic variants are defined, the method of the invention identifies the coordinates of the changes by mapping onto a reference genome the normal consensus sequences corresponding to each of the variants, avoiding potential mapping conflicts derived from the presence of the variant, as usually happens when using reference-based approaches. Sequences mapping (with the same score) to several positions in the genome are discarded. The same process can be done by mapping onto virus databases to locate which are the viruses inserted in the genome being analyzed. Construction of the in silico chromosome 20

In order to measure and calibrate the detection capabilities (sensitivity) of the method of the invention, inventors executed it on a controlled system, consisting of modified sequences of the chromosome 20. A personalized chromosome 20 was extracted from the hg19 reference genome downloaded from UCSC (with no repeat-masking)

(http://www.ucsc.edu) and modified to match a randomly chosen human haplotype. This chromosome contains 148,639 variants consisting of 96,935 SNPs and 51 ,704 deletions. The catalogue of somatic variants further added to this personalized chromosome and constituting the target of the invention, includes random 168 SNVs, 12 random deletions, 18 random insertion, 6 inversions and 1 insertion of Kl polyomavirus (extracted from:

http://www.ncbi.nlm.nih.qov/Taxonomy/Browser/wwwtax.cqi?l vl=0&id=423445 )

In silico sequencing was simulated using ART lllumina (Huang W. et.al. "ART: a next-generation sequencing read simulator" Bioinformatics 2012, vol. 28, pp. 593-594). For this inventors, like on the previous in silico, first generated a profile using the M0004 sample to extract parameters, like sequencing variation or read length.

Analysis of the in silico chromosome 20 with the method of the invention

The method of the invention was run with using the next variables X1 = 90, X2=28, X3=4, X4=1 , X5=7 and X6=10 considered as default at the moment of the invention.

For the sake of clarity and simplicity, inventors have set up X4 as 1 in this example of the in silico chromosome 20.

A System Configuration

The invention also comprised an efficient hardware configuration for the system focused on the execution of the computer program performing the method.

Due to the characteristics of the method, which involve a highly parallelizable input without dependencies and huge amounts of intermediate data, the input sources are split into chunks of C1 sequences and processed in a pipelined fashion. Thus, each step of the method is executed by a different component of the computer system, in parallel, over different chunks of input data. This strategy is more efficient in terms of execution time since it attempts to maximize the utilization of the system, and also overlaps data movement times.

When accelerators are capable of streaming data while computing, it is possible to add additional steps to the pipeline, eg. data is being sent to and received from the accelerators. When using accelerators that have

interconnections unable to offer bidirectional concurrent transfers, data movement must be serialized and this could become the bottleneck of the entire pipeline. Generally, the bigger the number of sequences per chunk (C1 ), the higher the achievable bandwidth on data transfers from host system to accelerators and vice versa is. On the other hand, this number is limited by the memory capacity of the accelerators. Moreover, encoding the keys may require to sorting all keys and some parallel sorting algorithms require a number of items that's a power 2, meaning extra padding might be required in order to reach the next power of 2. When this happens, the number of reads sent to each accelerator must be accordingly chosen in order to minimize the number of pudding elements.

The hardware components of the computer system consisted of:

A) Input Source

As input source, the processing pipeline reads high quality sequencing data from local disks or from network resources. An example of a network resource, is a sequencing machine connected to the system. In this case the system receives the input reads directly from the sequencing machine and starts elaborating at the same time as the sequencing machine is still working.

B) Host System, and C) Accelerators

In order to compute and elaborate data, the invention comprises processors and accelerators for example, hardware in the form of PCI cards or network resources, for offloading part of the computation. The amount of work to offload to the accelerators depends on the interconnection between

accelerator and host system as well on the processing power and memory of both processors and accelerators.

Ideally, processors will read the input sources that might need to be

uncompressed and filtered in order to extract only sequence reads and quality markers. Once enough sequences to fill a chunk are loaded in memory, sequences and relative quality markers are sent to the accelerators. In turn, accelerators will: filter reads, convert all read fragments to their encoded key representation; reduce k-mers producing the <key, count> pairs; and eventually, hashing the keys to obtain the tuple <key, count, hash>. Once the data is transferred back to the system's main memory, processors will consume it updating the counters stored in the associative structure.

Optionally, when accelerators have network interconnections and inputs are streamed from network resources, accelerators can read input directly from the source without requiring any work from the main system, offloading even more work. When the capabilities of the accelerators are limited, some of the processing steps can be carried out by the processors instead. For example, if transferring the tuple <key, count, hash> from accelerators to the host system becomes the bottleneck of the processing pipeline, it's wise to migrate the hashing step to the processors. In this way, it is reduced to 3/5 the amount of data that is transferred from the accelerators to the host system.

If the steps offloaded to the accelerators constitutes the bottleneck and having a higher number of updates to the hash table does not have an impact on the performance, the key conversion step can be disabled in order to increase the global throughput.

If more than one accelerator is available, data is split in different parts and each one of the parts is sent to a different accelerator. In order to efficiently offload the computation among multiple accelerators, the criterion used to split the data may take into account the available memory of each accelerator and may be as balanced as possible to the throughput offered by each accelerator. Which, for example, is: if two accelerators are available and one of them offers a throughput that is double of the other then the former accelerator should process an amount of data that is about 2/3 of the total, meanwhile the latter accelerator is about 1/3. If accelerators are different among each other and it's clear that different accelerators might suit better for different steps; then, the steps of the method are split among the available accelerator assigning each step to the accelerator that suits better. In this case, the output of one accelerator becomes the input of another and if no direct interconnection is available between the accelerators the host system must intervene to orchestrate data transfers. On the other hand, if no accelerators are available all the computation must be carried out by the main processors.

D) Main Memory, and E) Memory Expansion Cards

In addition to staging intermediate data from a one step of the pipeline to the next one, main memory is also used to store a partial version of the

associative structure. This in-memory structure is used to store k-mers seen just few times, while those seen more than 11 times are stored into a permanent associative structure in the memory expansions card.

In order to store the persistent associative structure containing all the useful information from normal (CNR) and tumoral (CTR) sample, memory expansion cards are used. When multiple cards are available inventors can use each card to store part of the associative structure allowing to split read requests among the cards increasing the possible number of in-flight requests proportionally with the number of the cards. Examples of an apparatus that can be used as memory expansion cards may be NVMe cards.

FIG. 1 shows the data flow from the input source, first to host system to load the reads, and then to accelerators for quality checking and base conversion. Afterwards, data is reduced again by the accelerators, which also generate the hash before data is loaded back into memory.

Results As explained above, to assess the performance of the method of the invention, inventors measured the fraction of somatic variants detected (sensitivity).

In silico validation with chromosome 20

This sample was created exclusively in order to validate the method against an in silico sample with an insertion of a virus, also the sample includes different somatic mutations.

Inventors first observed that the calling of somatic SNVs and SVs is optimal with a sensitivity of 96.4% for SNVs, 96.2% for Small SVs, 100% for Large SVs and 100% for virus Insertion. It is remarkable the capacity of detecting large structural variants with such a high sensitivity (Table 3)

Table 3. Assessment variant calling for chromosome 20 FIG. 2 provides an overview of the filtering capabilities of the invention, along with its sensitivity, at different stages of the execution. Step D) (filter) reduces the number of input reads to less than 10% of the input, while still keeping 99% of mutations identifiable, and high number of reads supporting those mutations (72%). Step E) (clustering) further decreases the size of relevant reads, while still keeping very high number of identifiable mutations.

Besides that, it is also capable of detecting the insertion of the Kl

polyomavirus. The computer-implemented method of the invention allows finding viral integration events with better accuracy and recall than the available alternative methods which are based on pre-alignment steps of the reads onto a reference genome. (FIG. 3). In the case of normal reads, inventors observed the genome sequence without any kind of structural variation. In contrast, on tumor reads (with heterozygosity A1 -A2) inventors were able to detect the Kl polyomavirus insertion on chromosome 20 at position 56398701 . In silico validation of the chromosome 20 insertion with the method disclosed in Moncunill et. al (ibid):

This method, with the default parameters, performed poorly in this test.

Its results on large structural variants does not show any evidence of the detection of Kl polyomavirus insertion on chromosome 20 at position

56398701 , it was just able to describe variants from the own genome.

Therefore, the method of the present invention was shown to be superior to the reference-free suffix-tree- based method described in Moncunill et al.

Taking into account all the results obtained at this point, the method of the invention has shown the capacity to detect all kinds of SNVs and SVs with great sensitivity without restrictions with the size. Also it is capable to detect the insertions of a virus in the genome with a base-pair resolution in comparison with the available alternative methods, which are based on suffix trees.

REFERENCES CITED IN THE APPLICATION Li, H., et.al. "Fast and accurate short read alignment with Burrows-Wheeler transform" Bioinformatics 2009, vol. 25, pp. 1754-1760 Medvedev P. "Computational methods for discovering structural variation with next-generation sequencing" Nat. Methods Suppl. 2009, vol. 6, S13-S20

Ding, L, et.al. "Expanding the computational toolbox for mining cancer genomes" Nat. Rev. Genet. 2014, vol. 15, pp. 556-570

Chen K. "TIGRA: A targeted Iterative Graph Routing Assembler for

Breakpoint Assembly" Genome Res. 2016, vol. 24, pp.310-317.

Zhuang, J. "Local sequence assembly reveals a high-resolution profile of somatic structural variations in 97 cancer genomes" Nucleic Acid Research 2015, vol. 43, pp.8146-8156

Moncunill V., et al., "Comprehensive characterization of complex structural variations by directly comparing genome sequence reads" Nature Biotech. 2014, vol. 32, pp. 1 106-1 1 12

Patro R., et. al. "Sailfish enables alignment-free isoform quantification from RNA-seq reads using lightweight algorithms" Nature Biotech. 2014, vol. 32, pp. 462-464 Nordstrom K.J.V, et.al. "Mutation identification by direct comparison of whole- genome sequencing data from mutant and wild-type individuals using k-mers" Nature Biotech. 2013, vol. 31 , pp. 325-331

Paszkiewicz K., et.al. "De novo assembly of short sequence reads" Brief Bioinform. 2010, vol. 1 1 , pp. 475-472

Huang W. et.al. "ART: a next-generation sequencing read simulator"

Bioinformatics 2012, vol. 28, pp. 593-594