Publications

2026

• D-LIM: A neural network for interpretable gene–gene interactions
  
  • Wang Shuhui
  • Allauzen Alexandre
  • Nghe Philippe
  • Opuu Vaitea
  PLoS Computational Biology, PLOS, 2026. <div><p>Recent advances in gene editing can produce large genotype-fitness maps for targeted genes, yet predicting the effects of mutations between genes remains challenging. Indeed, biochemical models require knowledge of underlying parameters and interactions, whereas machine learning methods typically lack interpretability, as they do not link model parameters to biological quantities. We introduce D-LIM, a neural network that infers low-dimensional fitness landscapes directly from mutation-fitness data. The distinctive feature of D-LIM is that it assumes genes act through independent gene-specific molecular phenotypes whose nonlinear interactions determine fitness. When this assumption holds, the model yields accurate predictions and interpretable effective phenotypes. Conversely, failure reveals that a low-dimensional model is insufficient. Applied to deep mutational scanning of metabolic pathways, protein-protein interactions, and yeast environmental adaptation, D-LIM achieves state-ofthe-art predictive accuracy. The inferred phenotype-fitness landscapes reveal whether epistatic interactions can be captured by a low-dimensional continuous model and identify potential trade-offs. Moreover, D-LIM estimates mutational effects on the effective phenotypes, enabling weak extrapolation beyond the training domain. D-LIM demonstrates how simple structure constraints in a neural network can help inference and hypothesis generation in biology.</p></div>
• Autocatalytic cores in the diluted regime: classification and properties
  
  • Nandan Praneet
  • Nghe Philippe
  • Unterberger Jérémie
  Journal of Mathematical Biology, Springer, 2026, 92 (3), pp.36. Autocatalysis underlies the ability of chemical and biochemical systems to replicate. Autocatalysis was recently defined stoichiometrically for reaction networks; five types of minimal autocatalytic networks, termed autocatalytic cores were identified. A necessary and sufficient stoichiometric criterion was later established for dynamical autocatalysis in diluted regimes, ensuring a positive growth rate of autocatalytic species starting from infinitesimal concentrations, given that degradation rates are sufficiently low. Here, we show that minimal autocatalytic networks in the dynamical sense, in the diluted regime, follow the same classification as autocatalytic cores in the stoichiometric sense. We further prove the uniqueness of the stationary regimes of autocatalytic cores, with and without degradation, for all types, except types II with three catalytic loops or more, for which the question remains open. These results indicate that the stationary point is robust under perturbation at low concentrations. More complex behaviours require additional non-linear couplings. (10.1007/s00285-026-02357-7)
  DOI : 10.1007/s00285-026-02357-7
• Biochemical insights into the conserved interactions of NMD factors from budding yeast to humans
  
  • Ruiz-Gutierrez Nadia
  • Graille Marc
  • Le Hir Hervé
  • Saveanu Cosmin
  , 2026. Nonsense-mediated mRNA decay (NMD) is one of the most extensively studied pathways of cytoplasmic mRNA degradation. It plays a critical role in diverse cellular processes by eliminating aberrant transcripts containing premature stop codons and by regulating the stability of physiological mRNAs. NMD factors were initially identified through genetic screens in S. cerevisiae (UPF1, 2, 3) and C. elegans (SMG-1, SMG5-7). Subsequent biochemical studies revealed the composition of NMD complexes and identified additional factors. A major protein hub for NMD is Upf1, an ATP-dependent RNA helicase that is part of two mutually exclusive NMD assemblies, the Upf1-Upf2-Upf3 complex and the Upf1-decapping complex, which contains the decapping enzyme and its co-factors. Here, we discuss recent findings, primarily from budding yeast, on the protein-protein interactions driving NMD complex dynamics and their similarities to human NMD. Together, the N-terminal cysteine and histidine rich (CH) and helicase domains (HD) of Upf1 act as a hub for binding multiple partners. Upf1 is required for binding to NMD substrates and for the initiation of RNA degradation through decapping (yeast) or endonucleolytic hydrolysis (humans). We focus on the interplay between Upf2, Dcp2 and Nmd4 (yeast SMG6), which ensures the mutually exclusive formation of Upf1-bound subcomplexes modulating Upf1's affinity for RNA. Thus, the study of NMD factors interactions in different organisms sheds new light on the remarkable conservation of NMD molecular mechanisms.
• Deux nouveaux acteurs de la libération contrôlée des vésicules extracellulaires
  
  • Bizingre Chloé
  • Picard Flavien
  • Alleaume-Butaux Aurélie
  • Pietri Mathéa
  • Baudry Anne
  • Schneider Benoit
  Médecine/Sciences, EDP Sciences, 2026, 42 (1), pp.25-28. (10.1051/medsci/2025245)
  DOI : 10.1051/medsci/2025245
• The Logic of Proof of Concept Research
  
  • Malaterre Christophe
  • Nghe Philippe
  Mind, Oxford University Press (OUP), 2026. Proof of Concept Research (PoCR) is a prevalent facet of scientific inquiry, yet its epistemic features remain poorly understood. While novelty has been highlighted as a key characteristic, projectability—understood as the likelihood of being applicable to a broader range of contexts—is another. This study endeavours to construct a formal model that elucidates the implicit ampliative reasoning inherent in PoCR. Our model hinges on probability assumptions for target objects to simultaneously exhibit three properties: one that is a defining characteristic of these target objects; a second that is desired of them and whose demonstration is the empirical aim of PoCR; and a third that is promised in the background. Depending on assumptions about when these properties jointly obtain, we delineate paradigmatic, alternative, and tangential modes of reasoning. This classification and associated decision tree unveil distinct argumentative strategies that, despite not being deductively valid, may be employed to motivate PoCR and justify subsequent inferences upon successful proof of concept demonstration. The model and decision tree together provide a framework with which to better understand the general structure of widely used inferences in PoCR, and with which researchers and evaluators can more precisely design and assess PoCR projects. (10.1093/mind/fzaf051)
  DOI : 10.1093/mind/fzaf051