Approximate membership query (AMQ) data structures are widely used for indexing the presence of elements drawn from a large set. To represent the count of indexed elements, AMQ data structures can be generalized into "counting AMQ" data structures. This is for instance the case of the "counting Bloom filters". However, counting AMQ data structures suffer from false positive and overestimated calls. In this work we propose a novel strategy, called fimpera, that reduces the false-positive rate and overestimation rate of any counting AMQ data structure indexing k-mers (words of length k) from a set of sequences, along with their abundance. Applied on a counting Bloom filter, fimpera decreases its false-positive rate by an order of magnitude while reducing the number of overestimated calls. Furthermore, fimpera lowers the average difference between the overestimated calls and the ground truth. In addition, it slightly decreases the query time. fimpera does not require any modification of the original counting AMQ data structure, it does not generate false-negative calls, and causes no memory overhead. The unique drawback is that fimpera yields a negligible amount of a new kind of false positives and overestimated calls 18.

Public sequencing databases contain vast amounts of biological information, yet they are largely underutilized as it is challenging to efficiently search them for any sequence(s) of interest. We developped kmindex (7.1.2), an approach that can index thousands of metagenomes and perform sequence searches in a fraction of a second. The index construction is an order of magnitude faster than previous methods, while search times are two orders of magnitude faster. With negligible false positive rates below 0.01%, kmindex outperforms the precision of existing approaches by four orders of magnitude. We demonstrate the scalability of kmindex by successfully indexing 1,393 marine seawater metagenome samples from the Tara Oceans project. Additionally, we introduce the publicly accessible web server “Ocean Read Atlas”, which enables real-time queries on the Tara Oceans dataset 37.

Comprehensive collections approaching millions of sequenced genomes have become central information sources in the life sciences. However, the rapid growth of these collections makes it effectively impossible to search these data using tools such as BLAST and its successors. To address this challenge, we developed a technique called phylogenetic compression, which uses evolutionary history to guide compression and efficiently search large collections of microbial genomes using existing algorithms and data structures 32. We showed that, when applied to modern diverse collections approaching millions of genomes, lossless phylogenetic compression improves the compression ratios of assemblies, de Bruijn graphs, and $k$-mer indexes by one to two orders of magnitude (implemented in a software tool called MiniPhy (7.1.10)). Additionally, we developed a pipeline called MOF-Search (7.1.9) for a BLAST-like search over these phylogeny-compressed reference data, and demonstrated it can align genes, plasmids, or entire sequencing experiments against all sequenced bacteria until 2019 on ordinary desktop computers within a few hours. Phylogenetic compression has broad applications in computational biology and may provide a fundamental design principle for future genomics infrastructure 32.

Efficiently managing large DNA datasets necessitates the development of highly effective sequence compression techniques to reduce storage and computational requirements. Here, we explore the potential of a lossy compression technique, Mapping-friendly Sequence Reductions (MSRs) which is a generalization of homopolymer compression to improve the accuracy of alignment tools. Essentially, MSRs deterministically transform sequences into shorter counterparts, in such a way that if an original query and a target sequence align, their reduced forms will align as well. While homopolymer compression is one example of an MSR, numerous others exist. These rapid computations yield lossy representations of the original sequences. Notably, the reduced sequences can be stored, aligned, assembled, and indexed much like regular sequences. MSRs could be used to improve the efficiency of taxonomic classification tools, by indexing and querying reduced sequences. Our experimentation with a mixture of 10 E. coli strains, demonstrates that this approach can yield greater precision than indexing and querying a reduced portion of k-mers. Other tasks could benefit from sequence reduction, such as mapping, genome assembly, and structural variant detection 26.

Despite the implementation of strict laboratory protocols for controlling ancient DNA contamination, samples still remain highly vulnerable to environmental contamination. Such contamination can significantly alter microbial composition, leading to inaccurate conclusions in downstream analyses. Within the co-supervision of a PhD student at the Institut Pasteur (Paris, France), we contributed to the work related to two $k$-mer-based methods for addressing the following two challenges in ancient metagenomics: (i) microbial source tracking for contamination assessment 13 and (ii) read-level contamination removal 12.

The first method is based on the construction of a k-mer matrix 8 which stores the presence/absence of k-mers across multiple samples of different well-characterised sources. Such a matrix is then used to predict the proportion of each source in unknown input samples.

The second method allows to retain, from an input contaminated set, reads likely to belong to a specific source of interest. On synthetic data, it achieves over 89.53% sensitivity and 94.00% specificity. On real datasets, aKmerBroom shows higher read retainment (+60% on average) than competing methods.

Squares (fragments of the form $xx$, for some string $x$) are arguably the most natural type of repetition in strings. The basic algorithmic question concerning squares is to check if a given string of length n is square-free, that is, does not contain a fragment of such form. We show that testing square-freeness of a length-n string over general alphabet of size $\sigma$ can be done with $O(nlog\sigma )$ comparisons, and cannot be done with $o(nlog\sigma )$ comparisons. This result comes with an $O(nlog\sigma )$ time algorithm in the Word RAM model 21.

The fundamental question considered in algorithms on strings is that of indexing, that is, preprocessing a given string for specific queries. By now we have a number of efficient solutions for this problem when the queries ask for an exact occurrence of a given pattern $P$. However, practical applications motivate the necessity of considering more complex queries, for example concerning near occurrences of two patterns.

Recently, Bille et al.  45 introduced a variant of such queries, called gapped consecutive occurrences, in which a query consists of two patterns $P_1$ and $P_2$ and a range $[a,b]$, and one must find all consecutive occurrences ($q_1,q_2$) of $P_1$ and $P_2$ such that $q_2-q_1\in [a,b]$. By their results, we cannot hope for a very efficient indexing structure for such queries, even if $a=0$ is fixed (although at the same time they provided a non-trivial upper bound). Motivated by this, we focus on a text given as a straight-line program (SLP) and design an index taking space polynomial in the size of the grammar that answers such queries in time optimal up to polylog factors 22, 35.

The popularity of $k$-mer-based methods has recently led to the development of compact k-mer-set representations, such as simplitigs/Spectrum-Preserving String Sets (SPSS), matchtigs, and eulertigs. These aim to represent $k$-mer sets via strings that contain individual $k$-mers as substrings more efficiently than the traditional unitigs. We demonstrated that all such representations can be viewed as superstrings of input $k$-mers, and as such can be generalized into a unified framework that we call the masked superstring of $k$-mers 39. We studied the complexity of masked superstring computation and proved NP-hardness for both $k$-mer superstrings and their masks 39. We then designed local and global greedy heuristics for efficient computation of masked superstrings, implemented them in a program called KmerCamel (7.1.11), and evaluate their performance using selected genomes and pan-genomes 39. Overall, masked superstrings unify the theory and practice of textual $k$-mer set representations and provide a useful framework for optimizing representations for specific bioinformatics applications.

Scaffolding is an intermediate stage of fragment assembly. It consists in orienting and ordering the contigs obtained by the assembly of the sequencing reads. In the general case, the problem has been largely studied with the use of distances data between the contigs. Here we focus on a dedicated scaffolding for the chloroplast genomes. As these genomes are small, circular and with few repeats, numerous approaches have been proposed to assemble them. However, their specificities have not been sufficiently exploited. We give a new formulation for the scaffolding in the case of chloroplast genomes as a discrete optimisation problem, that we prove to be NP-Complete. It does not require distance information. It is focused on a genomic regions view, with the priority on scaffolding the repeats first. In this way, we encode the multimeric forms issue in order to retrieve several genome forms that can exist in the same chloroplast cell. In addition, we provide an Integer Linear Program (ILP) to obtain exact solutions that we implement in Python3 package khloraascaf. We test it on synthetic data to investigate its performance behaviour and its robustness against several chosen difficulties. While the scaffolding problem is traditionally defined with distances data, we show it is possible to avoid them in the case of the well-studied circular chloroplast genomes. The presented results show that the regions view seems to be sufficient to scaffold the repeats 34.

Local assembly consists in reconstructing a sequence of interest from a sample of sequencing reads without having to assemble the entire genome, which is time and labor intensive. This is particularly useful when studying a locus of interest, for gap-filling in draft assemblies, as well as for alternative allele reconstruction of large insertion variants. Whereas linked-read technologies have a great potential to assemble specific loci as they provide long-range information, while maintaining the power and accuracy of short-read sequencing, there is a lack of local assembly tools for linked-read data.

We present MTG-Link (7.1.6), a novel local assembly tool dedicated to linked-reads. The originality of the method lies in its read subsampling step which takes advantage of the barcode information contained in linked-reads mapped in flanking regions of each targeted locus. Our approach relies then on our tool MindTheGap  10 to perform local assembly of each locus with the read subsets. MTG-Link tests different parameters values for gap-filling, followed by an automatic qualitative evaluation of the assembly.

We validated our approach on several datasets from different linked-read technologies. We show that MTG-Link is able to successfully assemble large sequences, up to dozens of Kb. We also demonstrate that the read subsampling step of MTG-Link considerably improves the local assembly of specific loci compared to other existing short-read local assembly tools. Furthermore, MTG-Link was able to fully characterize large insertion variants in a human genome and improved the contiguity of a 1.3 Mb locus of biological interest in several individual genomes of the mimetic butterfly Heliconius numata 15.

Long read assemblers struggle to distinguish closely related strains of the same species and collapse them into a single sequence. This is very limiting when analysing a metagenome, as different strains can have important functional differences. We have designed a new methodology supported by a software called HairSplitter (7.1.15), which recovers the strains from a strain-oblivious assembly and long reads. The originality of the method lies in a custom variant calling step that works with erroneous reads and separates an unknown number of haplotypes. On simulated datasets, we show that HairSplitter significantly outperforms the state of the art when dealing with metagenomes containing many strains of the same species 25, 40

We also propose an alternative approach for the strain separation problem using Integer Linear Programming (ILP). We introduce a strain-separation module, strainMiner, and integrate it into an established pipeline to create strain-separated assemblies from sequencing data. Across simulated and real experiments encompassing a wide range of sequencing error rates (5-12%), our tool consistently compared favorably to the state-of-the-art in terms of assembly quality and strain reconstruction. Moreover, strainMiner substantially cuts down the computational burden of strain-level assembly compared to published software by leveraging the powerful Gurobi solver. We think the new methodological ideas presented in this paper will help democratizing strain-separated assembly 27.

One of the problems in Structural Variant (SV) analysis is the genotyping of variants. It consists in estimating the presence or absence of a set of known variants in a newly sequenced individual. Our team previously released SVJedi, one of the first SV genotypers dedicated to long read data. The method is based on linear representations of the allelic sequences of each SV. While this is very efficient for distant SVs, the method fails to genotype some closely located or overlapping SVs. To overcome this limitation, we present a novel approach, SVJedi-graph (7.1.5), which uses a sequence graph instead of linear sequences to represent the SVs.

In our method, we build a variation graph to represent in a single data structure all alleles of a set of SVs. The long reads are mapped on the variation graph and the resulting alignments that cover allele-specific edges in the graph are used to estimate the most likely genotype for each SV. Running SVJedi-graph on simulated sets of close and overlapping deletions showed that this graph model prevents the bias toward the reference alleles and allows maintaining high genotyping accuracy whatever the SV proximity, contrary to other state of the art genotypers. On the human gold standard HG002 dataset, SVJedi-graph obtained the best performances, genotyping 99.5% of the high confidence SV callset with an accuracy of 95% in less than 30 min 20.

A variation graph is a data structure that aims to represent variations among a collection of genomes. It is a sequence graph where each genome is embedded as a path in the graph with the successive nodes, along the path, corresponding to successive segments on the associated genome sequence. Shared subpaths correspond to shared genomic regions between the genomes and divergent path to variations: this structure features inversions, insertions, deletions and substitutions. The construction of a variation graph from a collection of chromosome-size genome sequences is a difficult task that is generally addressed using a number of heuristics such as those implemented in the state-of-the-art pangenome graph builders minigraph-cactus and pggb. The question that arises is to what extent the construction method influences the resulting graph and therefore to what extent the resulting graph reflects genuine genomic variations. We propose to address this question by constructing an edition script between two variation graphs built from the same set of genomes which provides a measure of similarity, and more importantly that enables to identify discordant regions between the two graphs. We proceed by comparing, for each genome, the two corresponding paths in the two graphs which correspond to two possibly different segmentations of the same genomic sequence. As such, for each interval defined by the nodes of the path of the genome in the first graph, we define a set of relations with the nodes of the second graph, such as equalities, prefix and suffix overlaps... which allows for a calculation of how many elementary operations, such as fusions and divisions of nodes, are required to go from one graph to another. We tested our method on variation graphs constructed using both simulated dataset as well as a real dataset made of 15 yeast telomere-to-telomere phased genome assemblies, with minigraph-cactus and pggb as the graph construction tools. We showed that two graphs built with the same tool, minigraph-cactus, but with different incorporation orders of genomes can be more different from one another than two graphs built with the two different tools. We also showed thar our distance allows to pinpoint and vizualize the specific areas of the graph and genomes that are impacted by the changes in segmentation. The method is implemented in a Python tool named Pancat (7.1.13) 33, 24.

Genomic regions under positive selection harbor variation linked for example to adaptation. Most tools for detecting positively selected variants have computational resource requirements rendering them impractical on population genomic datasets with hundreds of thousands of individuals or more. We have developed and implemented an efficient haplotype-based approach able to scan large datasets and accurately detect positive selection. We achieve this by combining a pattern matching approach based on the positional Burrows–Wheeler transform with model-based inference which only requires the evaluation of closed-form expressions. We evaluate our approach with simulations, and find it to be both sensitive and specific. The computational resource requirements quantified using UK Biobank data indicate that our implementation is scalable to population genomic datasets with millions of individuals. Our approach may serve as an algorithmic blueprint for the era of “big data” genomics: a combinatorial core coupled with statistical inference in closed form 16.

We have developed a method based on a dynamic sliding window encoding (DSWE) for storing encrypted data in a DNA form taking into account biological constraints and prohibited nucleotide motifs used for data indexing. Its originality is twofold. First, it takes advantage of variable length DNA codewords to avoid homopolymers longer than $N$ bases when encoding binary data. Second, it relies on a sliding window to prevent the creation of prohibited motifs of nucleotides, adding non-coding bases when necessary. Contrarily to existing schemes, scaling DSWE to high values of $N$ and of numbers of prohibited motifs is extremely simple. It is furthermore independent of the cryptosystem. We provide the theoretical information rate of our proposal for a given number of prohibited motifs and a maximum homopolymer length. Experiments show that in general, it offers much higher performances than existing schemes. 23.

In absence of DNA template, the ab initio production of long double-stranded DNA molecules of predefined sequences is particularly challenging. The DNA synthesis step remains a bottleneck for many applications such as functional assessment of ancestral genes, analysis of alternative splicing or DNA-based data storage. We worked on a fully in vitro protocol to generate very long double-stranded DNA molecule starting from commercially available short DNA blocks in less than 3 days. This innovative application of Golden Gate assembly allowed us to streamline the assembly process to produce a 24 kb long DNA molecule storing part of the Universal Declaration of Human rights and citizens. The DNA molecule produced can be readily cloned into suitable host/vector system for amplification and selection. 36.

In the framework of the European BioPIM project, we have studied (from a parallelism point of view) a number of data structures used extensively in genomics software to assess the benefits of implementing them on PIM architectures. The following data structures have been studied: bloom filters, Burrows–Wheeler transform and hash tables. A detailed report on the evaluations is currently available on the GenoPIM project website.

The programming model for processing-in-memory is not yet well defined. The question of automatic (or semi-automatic) parallelization remains open. The tools available for programming a complete application on a PIM-equipped architecture are relatively low-level. We are working on the design of a C++ programming environment to unify the programming of CPU and PIM memory processing.

Memory components based on PIM principles have been developed by UPMEM, a young startup founded in 2015. The company has designed an innovative DRAM processing unit (DPU), a RISC processor integrated directly into the memory chip, on the DRAM die. We are using a UPMEM PIM server equipped with 160 GB of PIM memory and 256 GB of legacy memory to carry out full-scale experiments on the implementation of several genomic software for long DNA comparison, bacterial genome comparison, protein sequence alignment, data compression and sorting algorithms.

Initial experiments show that, for certain applications, it is possible to achieve a 20-fold acceleration compared with standard multicore platforms.

We developed an automated pipeline, Mapler (7.1.14), to assess the performances of long read metagenome assemblers, with a special focus on tools dedicated to high fidelity long reads such as PacBio HiFi reads. It compares five assembly tools, namely metaMDBG, metaflye, Hifiasm-meta, opera-ms and miniasm, and uses various evaluation metrics such as reference-based, reference-free and binning-based evalutation metrics. We applied this pipeline on several real metagenome datasets of various complexities, including publicly available mock communities with well characterized species contents and tunnel culture soil metagenomes with high and unknown species complexities. We showed that the different evaluation metrics are complementary and that high fidelity long read data allows drastic improvements in the number of obtained good quality Metagenome Assembled Genomes (MAGs) with respect to low-fidelity long reads or short read data, with MetaMDBG out-performing other HiFi-dedicated assemblers. However, for soil metagenomes most reconstructed MAGs remain of low quality because of the very large number of species in these microbiomes and the probable under-sampling of this diversity by this type of sequencing 41.

In this book chapter, we review the different bioinformatics analyses that can be performed on metagenomics and metatranscriptomics data. We present the differences of this type of data compared to standard genomics data and highlight the methodological challenges that arise from it. We then present an overview of the different methodological approaches and tools to perform various analyses such as taxonomic annotation, genome assembly and binning and de novo comparative genomics 28.

Through its long term collaboration with INRAE IGEPP, and its support to the BioInformatics of Agroecosystems Arthropods platform, GenScale is involved in various genomic and transcriptomics projects in the field of agricultural research. We participated in the genome assembly and analyses of some major agricultural pests or their natural ennemies. In particular, we performed a genome-wide identification of lncRNAs associated with viral infection in the lepidopteran pest Spodoptera frugiperda 19 and participated in a detailed study of a genomic region linked to reproductive mode variation in the pea aphid 17.

In most cases, the genomes and their annotations were hosted in the BIPAA information system, allowing collaborative curation of various sets of genes and leading to novel biological findings 42.

In the framework of a former ANR project (SpecRep 2014-2019), we worked on de novo genome assembly of several ithomiine butterflies. Due to their high heterozygosity level and to sequencing data of various quality, this was a challenging task and we tested numerous assembly tools. Finally, this work led to the generation of high-quality, chromosome-scale genome assemblies for two Melinaea species, M. marsaeus and M. menophilus, and a draft genome of the species Ithomia salapia. We obtained genomes with a size ranging from 396 Mb to 503 Mb across the three species and scaffold N50 of 40.5 Mb and 23.2 Mb for the two chromosome-scale assemblies. Various genomics and comparative genomics analyses were performed and revealed notably independent gene expansions in ithomiines and particularly in gustatory receptor genes.

These three genomes constitute the first reference genomes for the ithomiine butterflies (Nymphalidae: Danainae), which represent the largest known radiation of Müllerian mimetic butterflies and dominate by number the mimetic butterfly communities. This is therefore a valuable addition and a welcome comparison to existing biological models such as Heliconius, and will enable further understanding of the mechanisms of mimetism and adaptation in butterflies 14.

In some species, the Y is a tiny chromosome but the dioecious plant Silene latifolia has a giant 550 Mb Y chromosome, which has remained unsequenced so far. We participated in a collaborative project that sequenced and obtained a high-quality male S. latifolia genome. We participated in particular in the comparative analysis of the sex chromosomes with outgroups, that showed that the Y is surprisingly rearranged and degenerated for a 11 MY-old system. Recombination suppression between X and Y extended in a stepwise process, and triggered a massive accumulation of repeats on the Y, as well as in the non-recombining pericentromeric region of the X, leading to giant sex chromosomes 38.