Section: New Results

HTS data processing

Genome Analysis Tool Box Optimization

Participants: C. Deltel, P. Durand, E. Drezen, D. Lavenier, C. Lemaitre, P. Peterlongo, G. Rizk

Among the GATB library, the kmer-counting procedure is one of the most useful building block to speed-up development of new NGS tools. It is the first step of many NGS tools developed with GATB : Leon, Bloocoo, MindTheGap, DiscoSnp, Simka, TakeAbreak. This procedure has been optimized to be less limited by disks I/O. It relies on the use of kmer minimizers that help quickly partition the whole set of kmers into compact subsets. The kmer-counting procedure has also been re-worked to be more versatile, it is now able to count separately many input files and allows easy parametrization of the output, from simple kmer-count to the creation of custom user-defined kmer measures. At the core of the GATB library is also the manipulation and traversal of the de Bruijn Graph. The implementation has been optimized, leading to graph traversal twice fast as before. We introduced a new type of bloom filters, that are specially optimized for the manipulation of kmers. In these bloom filters neighboring kmers in the graph are close together in the bloom filter bit array, leading to better data locality, less cache misses and better overall performance [38] .

NGS Data Compression

Participants: G. Benoit, E. Drezen, D. Lavenier, C. Lemaitre, G. Rizk

A novel reference-free method to compress data issued from high throughput sequencing technologies has been developed. Our approach, implemented in the LEON software, employs techniques derived from assembly principles. The method is based on a reference probabilistic de-Bruijn Graph, built de novo from the set of reads and stored in a Bloom filter. Each read is encoded as a path in this graph, by memorizing an anchoring kmer and a list of bifurcations. The same probabilistic de-Bruijn Graph is used to perform a lossy transformation of the quality scores allowing higher compression rates to be obtained without loosing pertinent information for downstream analyses. Leon was run on various real sequencing datasets (whole genome, exome, RNA-seq or metagenomics). In all cases, LEON showed higher overall compression ratios than state-of-the-art compression software. On a C. elegans whole genome sequencing dataset, LEON divided the original file size by more than 20 [16] .

Multistep global optimization approach for the scaffolding problem

Participants: R. Andonov, D. Lavenier, I. Petrov

Our overall goal here is to address the computational hardness of the scaffolding problem by designing faster algorithms for global optimization that combine the branch-and-bound method which is able to find the global optimum but is usually slow for accuracy, with the use of massive parallelism and exploiting the special properties of the data–for scalability. A new two step scaffolding modeling strategy is in development. It tries to break the problem complexity by first solving a graph containing only large unitigs building something that can be compared to a trustworthy genomic frame. In our preliminary works [40] we developed integer programming optimization models that have been successfully applied on synthetic data generated from small chloroplast genomes. For computation we uses the Gurobi optimization solver.

Mapping reads on graph

Participants: A. Limasset, C. Lemaitre, P. Peterlongo

Next Generation Sequencing (NGS) has dramatically enhanced our ability to sequence genomes, but not to assemble them. In practice, many published genome sequences remain in the state of a large set of contigs. Although many subsequent analyses can be performed, one may ask whether mapping reads on the contigs is as informative as mapping them on the paths of the assembly graph. We proposed a formal definition of mapping on a de Bruijn graph, analysed the problem complexity which turned out to be NP-complete, and provided a practical solution. We proposed a pipeline called GGMAP (Greedy Graph MAPping). Its novelty is a procedure to map reads on branching paths of the graph, for which we designed a heuristic algorithm called BGREAT (de Bruijn Graph REAd mapping Tool). For the sake of efficiency, BGREAT rewrites a read sequence as a succession of unitigs sequences. GGMAP can map millions of reads per CPU hour on a de Bruijn graph built from a large set of human genomic reads. Surprisingly, results show that up to 22% more reads can be mapped on the graph but not on the contig set. Although mapping reads on a de Bruijn graph is a complex task, our proposal offers a practical solution combining efficiency with an improved mapping capacity compared to assembly-based mapping even for complex eukaryotic data [43] .

Improving discoSnp features

Participants: C. Riou, C. Lemaitre, P. Peterlongo

NGS data enable to detect polymorphisms such as SNPs and indels. Their detection in NGS data is now a routine task. The main methods for their prediction usually need a reference genome. However, non-model organisms and highly divergent genomes such as in cancer studies are more and more investigated. The discoSnp tool has been successfully applied to predict isolated SNPs from raw read set(s) without the need of a reference genome. We improved discoSnp which became discoSnp++ [44] . DiscoSnp++ benefits from a new software design that reduces time and memory consumption, and from a new algorithmic design that detects all kinds of SNP and small indels, adds genotype information and outputs a VCF (Variant Calling Format) file. Moreover, when a reference genome may be used, discoSnp++ predictions are automatically mapped to this reference and the VCF file shows up location information of each prediction. This step also provides a way to filter out false predictions due to genomic repeats. Using discoSnp++ even when a reference is available has multiple advantages: it is several order of magnitude faster and uses much less memory. We are currently working in showing that it also provides better predictions than methods based on read mapping.

HLA genotyping

Participant: D. Lavenier

The human leukocyte antigen (HLA) system drives the regulation of the Human immune system. Genotyping the HLA genes involved in the immune system consists first in a deep sequencing of the HLA region. Next, a NGS analysis is performed to detects SNP variations from which correct haplotypes are computed. We have developed a fast method that outperforms standard approaches which, generally, require exhaustive database searches. Instead, the method extracts a few significant k-mers from all the haplotypes referenced in the HLA database. Each haplotype is then characterized by a small set of informative k-mers. By comparing these k-mer sets with the HLA sequencing data of a specific person, we can rapidly determine its HLA genotype.

Identification of long non-coding RNAs in insects genomes

Participant: F. Legeai

The development of high throughput sequencing technologies (HTS) has allowed researchers to better assess the complexity and diversity of the transcriptome. Among the many classes of non-coding RNAs (ncRNAs) that were identified during the last decade, long non-coding RNAs (lncRNAs) represent a diverse and numerous repertoire of important ncRNAs, reinforcing the view that they are of central importance to the cell machinery in all branches of life. Although lncRNAs have been involved in essential biological processes such as imprinting, gene regulation or dosage compensation especially in mammals, the repertoire of lncRNAs is poorly characterized for many non-model organisms [23] . In collaboration with the Institut de Génétique et de Développement de Rennes (IGDR) we participate in the development of a software for extracting long non coding RNA from high throughput data (https://github.com/tderrien/FEELnc ).

Data-mining applied to GWAS

Participants D. Lavenier, Pham Hoang Son

Discriminative pattern mining methods are powerful techniques for discovering variant combinations related to diseases. The aim is to find a set of patterns that occur with disproportion frequency in case-control data sets, and a real challenge is to select a complete set of variant combinations that are biologically significant. There are various measurement methods for evaluating the discriminative power of individual combination in two-class data sets. Our research activity on this topic attempts to compare the statistical discriminative power measurements in genetic case-control data sets in order to evaluate the effectiveness of detecting variants associated with diseases.