Biology subjects

Menegaux, R.

Publications and source records attributed to Menegaux, R..

4 recordsLinked to original sources

bioRxiv2026

Foundation Models Improve Perturbation Response Prediction

Predicting cellular responses to genetic or chemical perturbations has been a long-standing goal in biology. Recent applications of foundation models to this task have yielded contradictory results regarding their superiority over simple baselines. We conducted an extensive analysis of over 600 different models across various prediction tasks and evaluation metrics, demonstrating that while some foundation models fail to outperform simple baselines, others significantly improve predictions for both genetic and chemical perturbations. Furthermore, we developed and evaluated methods for integrating multiple foundation models for perturbation prediction. Our results show that with sufficient data, these models approach fundamental performance limits, confirming that foundation models can improve cellular response simulations. Code and Data: https://github.com/genbio-ai/foundation-models-perturbation

bioinformatics↗

bioRxiv2025

Phylo-HS: A phylogenetic hierarchical softmax for taxonomic classification

Taxonomic binning -assigning taxonomic labels to DNA sequencing reads - is a core component of metagenomics data analysis. While machine learning approaches offer competitive speed and accuracy, their scalability is hindered by the growing number of referenced genomes and species. A major bottleneck lies in the final softmax layer of neural network models, which computes probabilities and gradients for all outcome classes. To address this, we propose Phylo-HS, a hierarchical softmax method that leverages the taxonomic tree to group classes into meaningful clusters. Phylo-HS achieves an order of magnitude speed improvement on a dataset with 5,000 classes and improves classification accuracy compared to frequency-based hierarchical softmax methods. By integrating phylogenetic structure into the model, Phylo-HS effectively balances scalability and accuracy for large-scale metagenomic analysis.

bioinformatics↗

bioRxiv2020

Embedding the de Bruijn graph, and applications to metagenomics

Fast mapping of sequencing reads to taxonomic clades is a crucial step in metagenomics, which however raises computational challenges as the numbers of reads and of taxonomic clades increases. Besides alignment-based methods, which are accurate but computational costly, faster compositional approaches have recently been proposed to predict the taxonomic clade of a read based on the set of k-mers it contains. Machine learning-based compositional approaches, in particular, have recently reached accuracies similar to alignment-based models, while being considerably faster. It has been observed that the accuracy of these models increases with the length k of the k-mers they use, however existing methods are limited to handle k-mers of lengths up to k = 12 or 13 because of their large memory footprint needed to store the model coefficients for each possible k-mer. In order to explore the performance of machine learning-based compositional approaches for longer k-mers than currently possible, we propose to reduce the memory footprint of these methods by binning together k-mers that appear together in the sequencing reads used to train the models. We achieve this binning by learning a vector embedding for the vertices of a compacted de Bruijn graph, allowing us to embed any DNA sequence in a low-dimensional vector space where a machine learning system can be trained. The resulting method, which we call Brume, allows us to train compositional machine learning-based models with k-mers of length up to k = 31. We show on two metagenomics benchmark that Brume reaches better performance than previously achieved, thanks to the use of longer k-mers.

bioinformatics↗

bioRxiv2018

Continuous embeddings of DNA sequencing reads, and application to metagenomics

We propose a new model for fast classification of DNA sequences output by next generation sequencing machines. The model, which we call fastDNA, embeds DNA sequences in a vector space by learning continuous low-dimensional representations of the k-mers it contains. We show on metagenomics benchmarks that it outperforms state-of-the-art methods in terms of accuracy and scalability.

bioinformatics↗