EN FR
EN FR


Section: New Results

Axis 1: Genomics

Transcriptome profiling using Nanopore sequencing Our vision of DNA transcription and splicing has changed dramatically with the introduction of short-read sequencing. These high-throughput sequencing technologies promised to unravel the complexity of any transcriptome. Generally gene expression levels are well-captured using these technologies, but there are still remaining caveats due to the limited read length and the fact that RNA molecules had to be reverse transcribed before sequencing. Oxford Nanopore Technologies has recently launched a portable sequencer which offers the possibility of sequencing long reads and most importantly RNA molecules. In [28], we generated a full mouse transcriptome from brain and liver using such Oxford Nanopore device. As a comparison, we sequenced RNA (RNA-Seq) and cDNA (cDNA-Seq) molecules using both long and short reads technologies and tested the TeloPrime preparation kit, dedicated to the enrichment of full-length transcripts. Using spike-in data, we confirmed in [28] that expression levels are efficiently captured by cDNA-Seq using short reads. More importantly, Oxford Nanopore RNA-Seq tends to be more efficient, while cDNA-Seq appears to be more biased. We further showed that the cDNA library preparation of the Nanopore protocol induces read truncation for transcripts containing internal runs of T’s. This bias is marked for runs of at least 15 T's, but is already detectable for runs of at least 9 T's and therefore concerns more than 20% of the expressed transcripts in mouse brain and liver. Finally, we outlined that bioinformatic challenges remain ahead for quantifying at the transcript level, especially when reads are not full-length. Accurate quantification of repeat-associated genes such as processed pseudogenes also remains difficult, and we show in the paper that current mapping protocols which map reads to the genome largely over-estimate their expression, at the expense of their parent gene.

Genotyping and variant detection The amount of genetic variation discovered and characterised in human populations is huge, and is growing rapidly with the widespread availability of modern sequencing technologies. Such a great deal of variation data, that accounts for human diversity, leads to various challenging computational tasks, including variant calling and genotyping of newly sequenced individuals. The standard pipelines for addressing these problems include read mapping, which is a computationally expensive procedure. A few mapping-free tools were proposed in recent years to speed up the genotyping process. While such tools have highly efficient run-times, they focus on isolated, bi-allelic SNPs, providing limited support for multi-allelic SNPs, indels, and genomic regions with high variant density. To address these issues, we introduced Malva , a fast and lightweight mapping-free method to genotype an individual directly from a sample of reads [10]. Malva is the first mapping-free tool that is able to genotype multi-allelic SNPs and indels, even in high density genomic regions, and to effectively handle a huge number of variants such as those provided by the 1000 Genome Project. An experimental evaluation on whole-genome data shows that Malva requires one order of magnitude less time to genotype a donor than alignment-based pipelines, providing similar accuracy. Remarkably, on indels, Malva provides even better results than the most widely adopted variant discovery tools.

Still on the issue of SNP detection, in [25], we developed the positional clustering theory that (i) describes how the extended Burrows–Wheeler Transform (eBWT) of a collection of reads tends to cluster together bases that cover the same genome position, (ii) predicts the size of such clusters, and (iii) exhibits an elegant and precise LCP array based procedure to locate such clusters in the eBWT. Based on this theory, we designed and implemented an alignment-free and reference-free SNP calling method, and we devised a SNP calling pipeline. Experiments on both synthetic and real data show that SNPs can be detected with a simple scan of the eBWT and LCP arrays as, in agreement with our theoretical framework, they are within clusters in the eBWT of the reads. Finally, our tool intrinsically performs a reference-free evaluation of its accuracy by returning the coverage of each SNP. Based on the results of the experiments on synthetic and real data, we conclude that the positional clustering framework can be effectively used for the problem of identifying SNPs, and it appears to be a promising approach for calling other types of variants directly on raw sequencing data.

Finally, variant detection and various related algorithmic problems were extensively explored in the PhD of Leandro I. S. de Lima [2] defended in April 2019.

Bubble generator Bubbles are pairs of internally vertex-disjoint (s,t)-paths in a directed graph, which have many applications in the processing of DNA and RNA data such as variant calling as presented above. Listing and analysing all bubbles in a given graph is usually unfeasible in practice, due to the exponential number of bubbles present in real data graphs. In [4], we proposed a notion of bubble generator set, i.e., a polynomial-sized subset of bubbles from which all the other bubbles can be obtained through a suitable application of a specific symmetric difference operator. This set provides a compact representation of the bubble space of a graph. A bubble generator can be useful in practice, since some pertinent information about all the bubbles can be more conveniently extracted from this compact set. We provided a polynomial-time algorithm to decompose any bubble of a graph into the bubbles of such a generator in a tree-like fashion. Finally, we presented two applications of the bubble generator on a real RNA-seq dataset.

Genome assembly The continuous improvement of long-read sequencing technologies along with the development of ad-doc algorithms has launched a new de novo assembly era that promises high-quality genomes. However, it has proven difficult to use only long reads to generate accurate genome assemblies of large, repeat-rich human genomes. To date, most of the human genomes assembled from long error-prone reads add accurate short reads to further improve the consensus quality (polishing). In a paper to be submitted before the end of 2019 (with as main authors A. di Genova and M.-F. Sagot), we report the development of an algorithm for hybrid assembly, Wengan , and its application to hybrid sequence datasets from four human samples. Wengan implements efficient algorithms that exploit the sequence information of short and long reads to tackle assembly contiguity as well as consensus quality. We show that the resulting genome assemblies have high contiguity (contig NG50:16.67-62.06 Mb), few assembly errors (contig NGA50:10.9-45.91 Mb), good consensus quality (QV:27.79-33.61), high gene completeness (BUSCO complete: 94.6-95.1%), and consume few computational resources (CPU hours:153-1027). In particular, the Wengan assembly of the haploid CHM13 sample achieved a contig NG50 of 62.06 Mb (NGA50:45.91 Mb), which surpasses the contiguity of the current human reference genome (GRCh38 contig NG50:57.88 Mb). Because of its lower cost, Wengan is an important step towards the democratisation of the de novo assembly of human genomes. Wengan is available at https://github.com/adigenova/wengan.

On assembly still, although haplotype-aware genome assembly plays an important role in genetics, medicine and various other disciplines, the generation of haplotype-resolved de novo assemblies remains a major challenge. Beyond distinguishing between errors and true sequential variants, one needs to assign the true variants to the different genome copies. Recent work has pointed out that the enormous quantities of traditional NGS read data have been greatly underexploited in terms of haplotig computation so far, which reflects the fact that the methodology for reference independent haplotig computation has not yet reached maturity. We presented in [7] a new approach, called POLYploid genome fitTEr (Polyte ) for a de novo generation of haplotigs for diploid and polyploid genomes of known ploidy. Our method follows an iterative scheme where in each iteration reads or contigs are joined, based on their interplay in terms of an underlying haplotype-aware overlap graph. Along the iterations, contigs grow while preserving their haplotype identity. Benchmarking experiments on both real and simulated data demonstrate that Polyte establishes new standards in terms of error-free reconstruction of haplotype-specific sequences. As a consequence, Polyte outperforms state-of-the-art approaches in various relevant aspects, notably in polyploid settings.

Others Besides the above, we have also explored a proteogenomics workflow for the expert annotation of eukaryotic genomes [18], as well as a technology- and species-independent simulator of sequencing data and genomic variants [42].