Jean‐Philippe Vert

The paper introduces MixupVI, a deep generative model for reference-free deconvolution of bulk RNA-seq data using single-cell transcriptomics, and releases PyDeconv for benchmarking deconvolution algorithms.

This paper introduces Sunbear, a framework for integrating multi-condition and multi-modal single-cell profiles across time to infer temporal profiles and predict dynamic relationships in mouse embryonic development.

The paper introduces Sceptic, a supervised pseudotime analysis model that improves prediction accuracy over existing methods for scRNA-seq and single-nucleus image data.

LSMMD-MA scales multimodal data integration for single-cell genomics to handle up to a million cells per modality, improving on existing methods by two orders of magnitude.

The paper introduces a new model, Pastis-NB, for inferring 3D genome architecture from Hi-C data by addressing overdispersion in contact counts, showing improved accuracy and robustness compared to existing methods.

This paper introduces DEDAL, a deep learning model for protein sequence alignment that improves accuracy on remote homologs compared to existing methods.

The paper benchmarks scHPTM analysis pipelines to identify best practices for experimental design and data analysis.

Polarbear translates between single-cell modalities using semi-supervised learning, improving accuracy over existing methods and enabling cross-modality cell matching.

CloneSig jointly infers intra-tumor heterogeneity and mutational signature activity in bulk tumor sequencing data, outperforming existing methods in scenarios with varying mutational signature distributions across clones.

DeepConsensus improves PacBio HiFi read accuracy and assembly quality using a gap-aware transformer model.

The paper introduces new RC-equivariant architectures for DNA sequence analysis, showing they outperform existing models.

This review surveys recent machine learning approaches for analyzing single-cell omics data from 2019-2020, focusing on methods for dimensionality reduction, batch normalization, cell type classification, trajectory inference, and multimodal integration.

This paper introduces epiGWAS, a method for detecting epistasis in genome-wide association studies, showing improved identification of gene-gene interactions compared to existing approaches.

This study benchmarks dimensionality reduction methods for scRNA-seq by varying parameters and finds that some methods benefit from tuning while others perform well with defaults.

This paper introduces GRISLI, a method for inferring gene regulation networks from scRNA-seq data using linear differential equations and velocity inference, showing improved performance over existing methods.

The paper introduces Brume, a method that reduces memory usage for training compositional machine learning models with longer k-mers in metagenomics, achieving better performance on benchmarks.

ACCOST identifies significant changes in chromatin conformation between Hi-C experiments by accounting for the genomic distance effect, outperforming existing methods.

This study evaluates six methods for estimating tumor heterogeneity from single-sample WES data and finds significant method variability and limited prognostic value.

LOTUS predicts cancer driver genes by integrating mutation and interaction data, outperforming existing methods across multiple cancer types.

MMD-MA embeds cells measured by different single-cell assays into a shared latent space without cell or feature correspondence, recovering true pairings in simulations and a 61-cell expression/methylation co-assay.

This paper introduces a method to infer 3D diploid chromatin structures from Hi-C data and validates it on simulated and mouse chromosome X data.

The paper introduces fastDNA, a model that embeds DNA sequences into a vector space for metagenomics classification, achieving higher accuracy and scalability than existing methods.

This paper introduces a normalization method for Hi-C data that accounts for copy number variations, improving downstream analysis of genome organization.

This study examines changes in genome organization during the transmission stages of Plasmodium parasites and finds that three-dimensional genome structure is linked to transcriptional activity of gene families involved in pathogenesis and immune evasion.

The paper introduces a weighting strategy using a zero-inflated negative binomial model to improve bulk RNA-seq tools for single-cell RNA-seq data with zero inflation.

The paper introduces ZINB-WaVE, a method for improving low-dimensional representations of scRNA-seq data by accounting for dropouts and over-dispersion, outperforming PCA and ZIFA.

This paper introduces kernel multitask regression to predict compound toxicity on cell lines using chemical and genetic data, achieving strong performance on a benchmark challenge.

This study shows that signaling pathway activities are better predictors of breast cancer prognosis than gene-level features and identify additional genes contributing to prognosis beyond pathway analysis.