Biology subjects

Malbranke, C.

Publications and source records attributed to Malbranke, C..

7 recordsLinked to original sources

bioRxiv2026

Predicting host-pathogen interactions using a proteome-scale language model

ProteomeLM (Malbranke et al., 2025) is a proteome-scale language model trained on proteomes spanning the tree of life to reconstruct masked protein embeddings from proteome context within each species. Its attention coefficients capture protein-protein interactions without supervision. Here, we show that this capability extends to cross-species host-pathogen interactions (HPI) across ten human pathogen taxa spanning viruses and bacteria, and can be further improved with lightweight fine-tuning. We introduce ProteomeLM-HPI, a parameter-efficient adaptation via LoRA, trained on concatenated host-pathogen proteomes to reconstruct masked pathogen embeddings from host context. ProteomeLM-HPI involves two key design choices: asymmetric masking (pathogen-heavy masking) and blocked self-attention. Systematic ablations show that both choices contribute. To assess generalization, we introduce a strict cross-species benchmark enforcing pathogen-level hold-out and 40% sequence-identity filtering. On this benchmark, Proteome-HPI improves AUC on 9 out of 10 unseen pathogens.

bioinformatics↗

bioRxiv2025

ProteomeLM: A proteome-scale language model allowing fast prediction of protein-protein interactions and gene essentiality across taxa

Language models trained on biological sequences are advancing inference tasks from the scale of single proteins to that of genomic neighborhoods. Here, we introduce ProteomeLM, a transformer-based language model that uniquely operates on entire proteomes from species spanning the tree of life. ProteomeLM is trained to reconstruct masked protein embeddings using the whole proteomic context, yielding contextualized protein representations that reflect proteome-scale functional constraints. Notably, ProteomeLMs attention coefficients encode protein-protein interactions (PPI), despite being trained without interaction labels. Furthermore, it enables interactome-wide PPI screening that is substantially more accurate, and orders of magnitude faster, than amino-acid coevolution-based methods. We further develop ProteomeLM-PPI, a supervised model that combines ProteomeLM embeddings and attention coefficients to achieve state-of-the-art PPI prediction across benchmarks and species. Finally, we introduce ProteomeLM-Ess, a supervised gene essentiality predictor that generalizes across diverse taxa. Our results demonstrate the potential of proteome-scale language models for addressing function and interactions at the organism level. Significance statementPredicting protein interactions and functions is a key challenge in biology. Although deep learning-based language models are advancing the analysis of individual protein sequences and of genomic neighborhoods, they struggle to capture properties involving all the proteins expressed in a cell, such as protein-protein interactions (PPI) and gene essentiality. We present ProteomeLM, a language model that reasons on entire proteomes across diverse species. ProteomeLM captures PPI without supervision, and enables more accurate and faster screening of entire interactomes than current sequence-based approaches. ProteomeLM also delivers state-of-the-art supervised PPI prediction, and improves supervised prediction of gene essentiality compared to protein language models. These results demonstrate the potential of proteome-scale language models to reveal system-level organization and functional relationships.

bioinformatics↗

bioRxiv2025

Out-of-equilibrium noise facilitates inference from protein sequence data

Homologous proteins have similar three-dimensional structures and biological functions that shape their sequences. The resulting coevolution-driven correlations underlie methods from Potts models to AlphaFold, which infer protein structure and function from sequences. Using a minimal model, we show that fluctuating selection strength and the onset of new selection pressures improve coevolutionbased inference of structural contacts. Our conclusions extend to realistic synthetic data and to the inference of interaction partners. Out-of-equilibrium noise arising from ubiquitous variations in natural selection thus enhances, rather than hinders, the success of inference from protein sequences.

biophysics↗

bioRxiv2025

Rubisco is slow across the tree of life

Rubisco is the main gateway through which inorganic carbon enters the biosphere, catalyzing the vast majority of carbon fixation on Earth. This pivotal enzyme has long been observed to be kinetically constrained. Yet, this impression is based on kinetic measurements heavily focused on eukaryotic rubiscos, a rather conserved group of low genetic diversity. Moreover, the fastest rubiscos that we know of so far were found among the sparsely sampled prokaryotes. Could there be yet faster rubiscos among the uncharted regions of rubiscos phylogenetic diversity? Here, we perform a characterization of more than 250 rubiscos from a wide range of bacteria and archaea, thereby doubling the coverage of the diversity of this key enzyme. We assess the distribution of the carboxylation rates at saturating levels of CO2, and establish that rubisco is a relatively slow enzyme across the tree of life, never exceeding {approx}30 reactions per second at 30{degrees}C. We show that relatively faster subclades share similar evolutionary contexts, involving micro-oxygenic environments or a CO2 concentrating mechanism. Leveraging a simple machine learning model trained on this dataset, we predict the carboxylation rate for all {approx}68,000 sequenced rubisco variants found in nature to date. This study provides the largest and most diverse dataset of natural variants for an enzyme and their associated rates, establishing a solid benchmark for future efforts to predict catalytic rates from sequence data. SignificanceDiscovering a fast carboxylating rubisco has been a long-standing challenge in the scientific community, given its potential impact on sustainable food and fuel production. Yet, only a small fraction of rubiscos natural diversity has been kinetically characterized. Here, we present a large-scale kinetic survey covering the entire spectrum of rubiscos diversity found in nature. Focusing on genetic clusters with above-average rates, we show that rubiscos catalytic rate does not exceed {approx}30 reactions per second at 30{degrees}C. Supported by a machine-learning predictive model, we extend this finding to all sequenced natural variants. Our study provides the most comprehensive kinetic dataset for a single enzyme to date.

systems biology↗

bioRxiv2024

ProtMamba: a homology-aware but alignment-free protein state space model

MotivationProtein language models are enabling advances in elucidating the sequence-to-function mapping, and have important applications in protein design. Models based on multiple sequence alignments efficiently capture the evolutionary information in homologous protein sequences, but multiple sequence alignment construction is imperfect. ResultsWe present ProtMamba, a homology-aware but alignment-free protein language model based on the Mamba architecture. In contrast with attention-based models, ProtMamba efficiently handles very long context, comprising hundreds of protein sequences. It is also computationally efficient. We train ProtMamba on a large dataset of concatenated homologous sequences, using two GPUs. We combine autoregressive modeling and masked language modeling through a fill-in-the-middle training objective. This makes the model adapted to various protein design applications. We demonstrate ProtMambas usefulness for sequence generation, motif inpainting, fitness prediction, and modeling intrinsically disordered regions. For homolog-conditioned sequence generation, ProtMamba outperforms state-of-the-art models. ProtMambas competitive performance, despite its relatively small size, sheds light on the importance of long-context conditioning. AvailabilityA Python implementation of ProtMamba is freely available in our GitHub repository: https://github.com/Bitbol-Lab/ProtMamba-ssm.

bioinformatics↗

bioRxiv2023

Computational design of novel Cas9 PAM-interacting domains using evolution-based modelling and structural quality assessment

We present here an approach to protein design that enables to leverage (i) scarce functional information such as experimental data (ii) evolutionary information learned from a natural sequence variants and (iii) physics-grounded modeling. Using a Restricted Boltzmann Machine (RBM), we learned a sequence model of a protein family. We use semi-supervision to leverage available functional information during the RBM training. We then propose a strategy to explore the protein representation space that can be informed by external models such as an empirical force field method (FoldX). This method was applied to a domain of the Cas9 protein responsible for recognition of a short DNA motif. We experimentally assessed the functionality of 71 variants that were generated to explore a range of RBM and FoldX energies. We show how a combination of functional, structural and evolutionary information can identify functional variants with high accuracy. Sequences with as many as 50 differences (20% of the protein domain) to the wild-type retained functionality. Overall, 21/71 sequences designed with our method were functional. Interestingly, 6/71 sequences showed an improved activity in comparison with the original wild-type protein sequence. These results demonstrate the interest in further exploring the synergies between machine-learning of protein sequence representations and physics grounded modeling strategies informed by structural information.

bioinformatics↗

bioRxiv2021

Improving sequence-based modeling of protein families using secondary structure quality assessment

MotivationModeling of protein family sequence distribution from homologous sequence data recently received considerable attention, in particular for structure and function predictions, as well as for protein design. In particular, Direct Coupling Analysis, a method to infer effective pairwise interactions between residues, was shown to capture important structural constraints and to successfully generate functional protein sequences. Building on this and other graphical models, we introduce a new framework to assess the quality of the secondary structures of the generated sequences with respect to reference structures for the family. ResultsWe introduce two scoring functions characterizing the likeliness of the secondary structure of a protein sequence to match a reference structure, called Dot Product and Pattern Matching. We test these scores on published experimental protein mutagenesis and design dataset, and show improvement in the detection of non-functional sequences. We also show that use of these scores help rejecting non-functional sequences generated by graphical models (Restricted Boltzmann Machines) learned from homologous sequence alignments. AvailabilitySupplementary Materials, Data and Code available at https://github.com/CyrilMa/ssqa. Contactcyril.malbranke@ens.fr O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=141 SRC="FIGDIR/small/428964v1_ufig1.gif" ALT="Figure 1"> View larger version (38K): org.highwire.dtl.DTLVardef@38d327org.highwire.dtl.DTLVardef@aeba12org.highwire.dtl.DTLVardef@13a09a2org.highwire.dtl.DTLVardef@14e85fa_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗