Biology subjects

Laine, E.

Publications and source records attributed to Laine, E..

At least 19 recordsLinked to original sources

bioRxiv2026

Constructed Negatives Shape the Evaluation of Sequence-Based Protein--Protein Interaction Models

Many machine-learning tasks, including representation learning, rely on negative examples that are constructed rather than directly observed. This issue is especially important for protein--protein interaction (PPI) prediction, where experimentally established non-interactions are scarce. PPI models are therefore commonly trained and evaluated against sampled, unobserved protein pairs treated as negatives. We study how this construction affects model comparison. Keeping positives, protein-disjoint splits, class balance, and model selection fixed, we vary training and evaluation negatives independently across six constructions and four sequence-based models, three of which use protein language model (pLM) representations. Changing only the test negatives shifts AUPRC by up to 0.26, and the top-ranked architecture changes with the construction. Negatives matched to true partners in pLM representation space reduce all pLM-based models to near-chance performance, and shuffling partners within each class shows that the original pairing contributes little to AUROC. Constructed training negatives rarely outperform uniform sampling: on external data, they underperform it in 108 of 120 comparisons for the pLM-based models, with gains never exceeding 0.05 AUROC and losses up to 0.23. We release splits, negative-generation procedures, and evaluation code that allow training and evaluation negatives to be varied separately. Our results show that with constructed negatives, model comparisons are not just noisy but systematically sensitive to how negatives are defined, a methodological choice that is often left implicit.

bioinformatics↗

bioRxiv2026

phyloDS: phylogeny-aware differential splicing across species, beyond reference annotations

Alternative splicing diverges rapidly between species, yet its comparative analysis lags far behind that of gene expression: junction orthology is defined on a single reference genome, species are treated as independent samples, and detected shifts are rarely grounded in sequence-level evidence. We present phyloDS, an RNA-seq--driven framework that resolves all three limitations. phyloDS defines orthologous splice junctions on multiple sequence alignments of one-to-one orthologous genes, tests their differential usage with a regularized Ornstein--Uhlenbeck model, and cross-validates detected shifts against splice-site predictions made from genomic sequence alone. Applied to primate brain and rodent heart transcriptomes, phyloDS recovered up to 2.9-fold more shared orthologous splice junctions than a reference-based strategy and kept type~I error near nominal where competing methods inflate it more than two-fold. It identified 234 and 502 differentially used orthologous splice junctions in hominoids and mole-rats (86 and 141 unannotated), revealing a mitochondrial and contractile splicing signature in mole-rats.

bioinformatics↗

bioRxiv2025

ProteoCast: a web server to predict, validate, and interpret missense variant effects

Understanding how mutations affect protein function remains critical yet challenging, particularly for variants in clinical databases lacking experimental characterisation and for intrinsically disordered regions. Current computational approaches often operate as black boxes, providing predictions without sufficient transparency or quality assessment of the underlying data. Here we present ProteoCast, a user-friendly web server that predicts variant effects through evolutionary constraint analysis and structural context integration. ProteoCast provides a three-tier variant classification (impactful, mild, neutral) to help prioritise mutations for clinical interpretation and experimental validation. It incorporates multiple sequence alignment quality controls to ensure prediction reliability and flag positions with insufficient evolutionary information. Beyond single-variant classification, ProteoCast employs a novel segmentation approach based on mutational sensitivity to identify functional linear peptides in disordered regions. Interactive visualisations guide users through results interpretation, from variant-level predictions to protein-wide functional landscapes. Evaluation on 63,000 ClinVar variants demonstrates 77% sensitivity and 87% specificity for pathogenicity prediction, with performance maintained across species (85% accuracy on Drosophila lethal mutations). ProteoCast successfully identifies twice as many functional motifs in intrinsically disordered regions compared to conservation-based phylogenetic methods. Predictions can be tuned to specific conformations, such as bound forms in protein complexes, for improved accuracy and interpretability. With its transparent, unsupervised methodology and computational efficiency (minutes per protein), ProteoCast democratises access to variant effect prediction and functional site discovery for the broader research community. The web server is freely available at: https://proteocast.ijm.fr/. HighlightsO_LIProteoCast web server democratises variant effect prediction through a fast, transparent platform combining evolutionary analysis and structural context C_LIO_LIThree-tier variant classification enables prioritisation for clinical interpretation and genome editing experiments C_LIO_LINovel mutational sensitivity segmentation approach identifies functional peptides in disordered regions C_LIO_LIInteractive visualisations and input alignment quality controls guide interpretation C_LI

bioinformatics↗

bioRxiv2025

Proteome-wide prediction of the Functional Impact of Missense Variants with ProteoCast

Dissecting the functional impact of genetic mutations is essential to advancing our understanding of genotype-phenotype relationships and identifying therapeutic targets. Despite progress in sequencing and genome editing technologies, proteome-wide mutation effect prediction remains challenging. Here we show that evolutionary information alone enables accurate prediction of mutation effects across entire proteomes. ProteoCast is a scalable and interpretable computational method that leverages protein sequence conservation to classify genetic variants and identify functionally important protein sites. We apply ProteoCast to the complete Drosophila melanogaster proteome (22,000 isoforms, 300 million mutations) and validate it against nearly 400,000 natural and experimental variants. It correctly classifies 85% of known lethal mutations as functionally impactful versus 13-18% of population variants. ProteoCast-guided genome editing experiments confirm these predictions. Moreover, ProteoCast successfully identifies functionally important protein modification sites and binding motifs. ProteoCast provides a publicly available resource and deployable pipeline for studying gene function and mutations in any organism.

evolutionary biology↗

bioRxiv2024

SeaMoon: Prediction of molecular motions based on language models

How protein move and deform determines their interactions with the environment and is thus of utmost importance for cellular functioning. Following the revolution in single protein 3D structure prediction, researchers have focused on repurposing or developing deep learning models for sampling alternative protein conformations. In this work, we explored whether continuous compact representations of protein motions could be predicted directly from sequences, without exploiting 3D structures. SeaMoon leverages protein Language Model (pLM) embeddings as input to a lightweight convolutional neural network. We assessed SeaMoon against [~] 1 000 collections of experimental conformations exhibiting diverse motions. It predicts at least one ground-truth motion with reasonable accuracy for 40% of the test proteins. SeaMoon captures motions inaccessible to normal mode analysis, an unsupervised physics-based method relying solely on 3D geometry, and generalises to proteins without detectable sequence similarity to the training set. SeaMoon is easily retrainable with novel or updated pLMs.

bioinformatics↗

bioRxiv2024

Unlocking Protein Evolution Insights: Efficient and Interpretable Mutational Effect Predictions with GEMME

The Global Epistatic Model for predicting Mutational Effects (GEMME) is a computational method that reconstructs protein mutational landscapes using sequence data alone. In this article, we delve into the broader biological questions that GEMME can address beyond mere landscape reconstruction. We provide several examples to guide users in maximizing the utility of GEMME. Additionally, we discuss recent advancements that enhance GEMMEs predictive accuracy and extend its capabilities by integrating sequence data with structural information and allele frequency data, when available. These innovations enable GEMME to answer new biological questions and offer deeper insights into protein evolution and function.

bioinformatics↗

bioRxiv2024

VespaG: Expert-guided protein Language Models enable accurate and blazingly fast fitness prediction

Exhaustive experimental annotation of the effect of all known protein variants remains daunting and expensive, stressing the need for scalable effect predictions. We introduce VespaG, a blazingly fast missense amino acid variant effect predictor, leveraging protein Language Model (pLM) embeddings as input to a minimal deep learning model. To overcome the sparsity of experimental training data, we created a dataset of 39 million single amino acid variants from the human proteome applying the multiple sequence alignment-based effect predictor GEMME as a pseudo standard-of-truth. This setup increases interpretability compared to the baseline pLM and is easily retrainable with novel or updated pLMs. Assessed against the ProteinGym benchmark (217 multiplex assays of variant effect - MAVE - with 2.5 million variants), VespaG achieved a mean Spearman correlation of 0.48{+/-}0.02, matching top-performing methods evaluated on the same data. VespaG has the advantage of being orders of magnitude faster, predicting all mutational landscapes of all proteins in proteomes such as Homo sapiens or Drosophila melanogaster in under 30 minutes on a consumer laptop (12-core CPU, 16 GB RAM). AvailabilityVespaG is available freely at https://github.com/jschlensok/vespag. The associated training data and predictions are available at https://doi.org/10.5281/zenodo.11085958.

bioinformatics↗

bioRxiv2024

Explaining Conformational Diversity in Protein Families through Molecular Motions

Proteins play a central role in biological processes, and understanding their conformational variability is crucial for unraveling their functional mechanisms. Recent advancements in high-throughput technologies have enhanced our knowledge of protein structures, yet predicting their multiple conformational states and motions remains challenging. This study introduces Dimensionality Analysis for protein Conformational Exploration (DANCE) for a systematic and comprehensive description of protein families conformational variability. DANCE accommodates both experimental and predicted structures. It is suitable for analysing anything from single proteins to superfamilies. Employing it, we clustered all experimentally resolved protein structures available in the Protein Data Bank into conformational collections and characterized them as sets of linear motions. The resource facilitates access and exploitation of the multiple states adopted by a protein and its homologs. Beyond descriptive analysis, we assessed classical dimensionality reduction techniques for sampling unseen states on a representative benchmark. This work improves our understanding of how proteins deform to perform their functions and opens ways to a standardised evaluation of methods designed to sample and generate protein conformations.

bioinformatics↗

bioRxiv2023

Charting the alternative usage of protein repeated regions in evolution and its implication for protein interactions

Alternative splicing of repeats in proteins provides a mechanism for rewiring and fine-tuning protein interaction networks. In this work, we developed a robust and versatile method, ASPRING, to identify alternatively spliced protein repeats from gene annotations. ASPRING leverages evolutionary meaningful alternative splicing-aware hierarchical graphs to provide maps between protein repeats sequences and 3D structures. We re-think the definition of repeats by explicitly accounting for transcript diversity across several genes/species. Using a stringent sequence-based similarity criterion, we detected over 5,000 evolutionary conserved repeats by screening virtually all human protein-coding genes and their orthologs across a dozen species. Through a joint analysis of their sequences and structures, we extracted specificity-determining sequence signatures and assessed their implication in experimentally resolved and modelled protein interactions. Our findings demonstrate the widespread alternative usage of protein repeats in modulating protein interactions and open avenues for targeting repeat-mediated interactions. HighlightsO_LIRobust detection of alternatively used repeated protein regions in evolution C_LIO_LIThe approach relies on sequence similarity and identifies conserved signatures C_LIO_LIMapping of the repeats onto protein isoform 3D models predicted by AlphaFold C_LIO_LI5 000 repeats detected over the human coding fraction, about one third disordered C_LIO_LIAssessment of the structural coverage of their interactions with protein partners C_LI

bioinformatics↗

bioRxiv2022

Alignment-based protein mutational landscape prediction: doing more with less

The wealth of genomic data has boosted the development of computational methods predicting the phenotypic outcomes of missense variants. The most accurate ones exploit multiple sequence alignments, which can be costly to generate. Recent efforts for democratizing protein structure prediction have overcome this bottleneck by leveraging the fast homology search of MMseqs2. Here, we show the usefulness of this strategy for mutational outcome prediction through a large-scale assessment of 1.5M missense variants across 72 protein families. Our study demonstrates the feasibility of producing alignment-based mutational landscape predictions that are both high-quality and compute-efficient for entire proteomes. We provide the community with the whole human proteome mutational landscape and simplified access to our predictive pipeline. Significant statementUnderstanding the implications of DNA alterations, particularly missense variants, on our health is paramount. This study introduces a faster and more efficient approach to predict these effects, harnessing vast genomic data resources. The speed-up is possible by establishing that resource-saving multiple sequence alignments suffice even as input to a method fitting few parameters given the alignment. Our results opens the door to discovering how tiny changes in our genes can impact our health. They provide valuable insights into the genotype-phenotype relationship that could lead to new treatments for genetic diseases.

bioinformatics↗

bioRxiv2022

Deep Local Analysis deconstructs protein-protein interfaces and accurately estimates binding affinity changes upon mutation

AO_SCPLOWBSTRACTC_SCPLOWThe spectacular recent advances in protein and protein complex structure prediction hold promise for reconstructing interactomes at large scale and residue resolution. Beyond determining the 3D arrangement of interacting partners, modeling approaches should be able to unravel the impact of sequence variations on the strength of the association. In this work, we report on Deep Local Analysis (DLA), a novel and efficient deep learning framework that relies on a strikingly simple deconstruction of protein interfaces into small locally oriented residue-centered cubes and on 3D convolutions recognizing patterns within cubes. Merely based on the two cubes associated with the wild-type and the mutant residues, DLA accurately estimates the binding affinity change for the associated complexes. It achieves a Pearson correlation coefficient of 0.81 on more than 2 000 mutations, and its generalization capability to unseen complexes is higher than the state-of-the-art methods. We show that taking into account the evolutionary constraints on residues contributes to predictions. We also discuss the influence of conformational variability on performance. Beyond the predictive power on the effects of mutations, DLA is a general framework for transferring the knowledge gained from the available non-redundant set of complex protein structures to various tasks. For instance, given a single partially masked cube, it recovers the identity and physico-chemical class of the central residue. Given an ensemble of cubes representing an interface, it predicts the function of the complex. Source code and models are available at http://gitlab.lcqb.upmc.fr/DLA/DLA.git.

bioinformatics↗

bioRxiv2022

Deep Local Analysis estimates effects of mutations on protein-protein interactions

The spectacular advances in protein and protein complex structure prediction hold promises for the reconstruction of interactomes at large scale at the residue resolution. Beyond determining the 3D arrangement of interacting partners, modeling approaches should be able to sense the impact of sequence variations such as point mutations on the strength of the association. In this work, we report on DLA-mutation, a novel and efficient deep learning framework for accurately predicting mutation-induced binding affinity changes. It relies on a 3D-invariant description of local 3D environments at protein interfaces and leverages the large amounts of available protein complex structures through self-supervised learning. It combines the learnt representations with evolutionary information, and a description of interface structural regions, in a siamese architecture. DLA-mutation achieves a Pearson correlation coefficient of 0.81 on a large collection of more than 2000 mutations, and its generalization capability to unseen complexes is higher than state-of-the-art methods.

bioinformatics↗

bioRxiv2022

Functional mapping of N-terminal residues in the yeast proteome uncovers novel determinants for mitochondrial protein import.

N-terminal ends of polypeptides are critical for the selective co-translational recruitment of N-terminal modification enzymes. However, it is unknown whether specific N-terminal signatures differentially regulate protein fate according to their cellular functions. In this work, we developed an in-silico approach to detect functional preferences in cellular N-terminomes, and identified in S. cerevisiae more than 200 Gene Ontology terms with specific N-terminal signatures. In particular, we discovered that Mitochondrial Targeting Sequences (MTS) show a strong and specific over-representation at position 2 of hydrophobic residues known to define potential substrates of the N-terminal acetyltransferase NatC. We validated mitochondrial precursors as co-translational targets of NatC by selective purification of translating ribosomes, and found that their N-terminal signature is conserved in Saccharomycotina yeasts. Finally, systematic mutagenesis of the position 2 in a prototypal yeast mitochondrial protein confirmed its critical role in mitochondrial protein import. Our work highlights the hydrophobicity of MTS N-terminal residues and their modification by NatC as critical features for the definition of the mitochondrial proteome, providing a molecular explanation for mitochondrial defects observed in yeast or human NatC-depleted cells. Functional mapping of N-terminal residues thus has the potential to support the discovery of novel mechanisms of protein regulation or targeting.

genomics↗

bioRxiv2022

Deep Local Analysis evaluates protein docking conformations with locally oriented cubes

AO_SCPLOWBSTRACTC_SCPLOWWith the recent advances in protein 3D structure prediction, protein interactions are becoming more central than ever before. Here, we address the problem of determining how proteins interact with one another. More specifically, we investigate the possibility of discriminating near-native protein complex conformations from incorrect ones by exploiting local environments around interfacial residues. Deep Local Analysis (DLA)-Ranker is a deep learning framework applying 3D convolutions to a set of locally oriented cubes representing the protein interface. It explicitly considers the local geometry of the interfacial residues along with their neighboring atoms and the regions of the interface with different solvent accessibility. We assessed its performance on three docking benchmarks made of half a million acceptable and incorrect conformations. We show that DLA-Ranker successfully identifies near-native conformations from ensembles generated by molecular docking. It surpasses or competes with other deep learning-based scoring functions. We also showcase its usefulness to discover alternative interfaces. Availabilityhttp://gitlab.lcqb.upmc.fr/dla-ranker/DLA-Ranker.git

bioinformatics↗

bioRxiv2021

From complete cross-docking to partners identification and binding sites predictions

Proteins ensure their biological functions by interacting with each other. Hence, characterising protein interactions is fundamental for our understanding of the cellular machinery, and for improving medicine and bioengineering. Over the past years, a large body of experimental data has been accumulated on who interacts with whom and in what manner. However, these data are highly heterogeneous and sometimes contradictory, noisy, and biased. Ab initio methods provide a means to a "blind" protein-protein interaction network reconstruction. Here, we report on a molecular cross-docking-based approach for the identification of protein partners. We applied it to a few hundred of proteins, and we systematically investigated the influence of several key ingredients, such as the size and quality of the interfaces and the scoring function. We achieved some significant improvement compared to previous works, and a very high discriminative power on some specific functional classes. In addition, we assessed the ability of the approach to account for protein surface multiple usages, and we compared it with a sequence-based deep learning method. This work may contribute to guiding the exploitation of the large amounts of protein structural models now available toward the discovery of unexpected partners and their complex structure characterisation.

bioinformatics↗

bioRxiv2021

HOPMA: Boosting protein functional dynamics with colored contact maps

In light of the recent very rapid progress in protein structure prediction, accessing the multitude of functional protein states is becoming more central than ever before. Indeed, proteins are flexible macromolecules, and they often perform their function by switching between different conformations. However, high-resolution experimental techniques such as X-ray crystallography and cryogenic electron microscopy can catch relatively few protein functional states. Many others are only accessible under physiological conditions in solution. Therefore, there is a pressing need to fill this gap with computational approaches. We present HOPMA, a novel method to predict protein functional states and transitions using a modified elastic network model. The method exploits patterns in a protein contact map, taking its 3D structure as input, and excludes some disconnected patches from the elastic network. Combined with nonlinear normal mode analysis, this strategy boosts the protein conformational space exploration, especially when the input structure is highly constrained, as we demonstrate on a set of more than 400 transitions. Our results let us envision the discovery of new functional conformations, which were unreachable previously, starting from the experimentally known protein structures. The method is computationally efficient and available at https://github.com/elolaine/HOPMA and https://team.inria.fr/nano-d/software/nolb-normal-modes.

bioinformatics↗

bioRxiv2020

Assessing Conservation of Alternative Splicing with Evolutionary Splicing Graphs

Understanding how protein function has evolved and diversified is of great importance for human genetics and medicine. Here, we tackle the problem of describing the whole transcript variability observed in several species by generalising the definition of splicing graph. We provide a practical solution to building parsimonious evolutionary splicing graphs where each node is a minimal transcript building block defined across species. We show a clear link between the functional relevance, tissue-regulation and conservation of AS events on a set of 50 genes. By scaling up to the whole human protein-coding genome, we identify a few thousands of genes where alternative splicing modulates the number and composition of pseudo-repeats. We have implemented our approach in ThorAxe, an efficient, versatile, and robust computational tool freely available at https://github.com/PhyloSofS-Team/thoraxe. The results are accessible and can be browsed interactively at http://www.lcqb.upmc.fr/ThorAxe.

bioinformatics↗

bioRxiv2020

Predicting substitutions to modulate disorder and stability in coiled-coils

Coiled-coils are described as stable structural motifs, where two or more helices wind around each other. However, coiled-coils are associated with local mobility and intrinsic disorder. Intrinsically disordered regions (IDRs) in proteins are characterized by lack of stable secondary and tertiary structure under physiological conditions in vitro. They are increasingly recognized as important for protein function. However, characterizing their behaviour in solution and determining precisely the extent of disorder of a protein region remains challenging, both experimentally and computationally. In this work, we propose a computational framework to quantify the extent of disorder within a coiled-coil in solution and to help design substitutions modulating such disorder. Our method relies on the analysis of conformational ensembles generated by relatively short all-atom Molecular Dynamics (MD) simulations. We apply it to the phosphoprotein multimerisation domains (PMD) of Measles virus (MeV) and Nipah virus (NiV), both forming tetrameric left-handed coiled-coils. We show that our method can help quantify the extent of disorder of the C-terminus region of MeV and NiV PMDs, without requiring the input MD trajectory to actually sample the unfolded states of these regions. Moreover, this study provided a conceptual framework for the rational design of substitutions aimed at modulating the stability of the coiled-coils. By assessing the impact of four substitutions known to destabilize coiled-coils, we derive a set of rules to control MeV PMD structural stability and cohesiveness. We therefore design two contrasting substitutions, one increasing the stability of the tetramer and the other increasing its flexibility. Consequently, our method can be considered as a platform to reason about how to design substitutions aimed at regulating flexibility and stability.

bioinformatics↗