Malo, L. | Search bioRxiv

bioRxiv2026

Agent-driven Model Development for RNA 3D Structure Prediction

Large language model (LLM) agents have shown promise in driving scientific discovery, but their effectiveness in complex, real-world biological problems remains underexplored. We ask whether a general-purpose LLM agent can drive semi-autonomous development of a model for a genuinely hard biological problem, RNA 3D structure prediction. We designed a development loop where, under a fixed budget and with human supervision, the agent iteratively proposed, implemented, trained, and evaluated model changes. Over 297 iterations, the model evolved from a randomly-initialised baseline to QuickFold, an 8.9M-parameter folding trunk that matches the strongest open-source baselines (RhoFold+, NuFold) on lDDT and TM-score within noise on a held-out test set at a fraction of their inference cost. We frame this less as a new predictor than as a case study in feedback-driven, agent-led model development.

bioinformatics↗

bioRxiv2025

Evolutionary and geometric signatures reveal ligand-binding sites across proteomes

Identifying protein binding sites is central to drug discovery, yet many computational approaches still trade off precision, recall, or throughput when scaled. We introduce PickPocket, a deep learning model that fuses evolutionary information from protein sequences with geometric representations of structure to identify ligand-binding residues at proteome scale. By leveraging complementary sequence context and spatial neighborhoods, PickPocket generalizes across diverse protein families and ligand chemistries while operating at a recall-oriented setting with competitive precision. In benchmark evaluations it delivers strong residue-level recovery and, despite no explicit training on conformational switching, reliably identifies cryptic pockets in held-out structures, comparing favorably with specialized approaches. Applied across 356,711 proteins, the method nominates previously unannotated candidate sites enriched for functional signals and highlights tractable surface chemistry on therapeutically relevant targets. These results position evolutionary-geometric fusion as a practical foundation for large-scale site mapping that can shorten the path from structure to experiment and support hit discovery, mutagenesis design, and target assessment.

bioinformatics↗

bioRxiv2021

Transcriptome analysis of the effect of AHR on productive and unproductive pathways of in vitro megakaryocytopoiesis

Human CD34+ progenitors can be differentiated in vitro into proplatelet-producing megakaryocytes (MKs) within 17 days. During this time, four cell populations emerge, phenotypically defined as CD34+CD41+ on day 7 (D7) and CD34+CD41+CD9- on D10 and D14 - qualified as "productive" because they can differentiate into proplatelet-forming cells during the D14-D17 period - and CD34-CD41+ or CD34+CD41+CD9+ on day 10 - qualified as "unproductive" because they are unable to form proplatelets later. Coculture with mesenchymal stem cells, or the presence of the AHR antagonist SR1, boosts the productive pathway in two ways: firstly, it increases the yield of D10 and D14 CD34+CD41+CD9- cells and secondly, it greatly increases their ability to generate proplatelets; in contrast, SR1 has no noticeable effect on the unproductive cell types. A transcriptome analysis was performed to decipher the genetic basis of these properties. This work represents the first extensive description of the genetic perturbations which accompany the differentiation of CD34+ progenitors into mature MKs at a subpopulation level. It highlights a wide variety of biological changes modulated in a time-dependent manner and allows anyone, according to his/her interests, to focus on specific biological processes accompanying MK differentiation. For example, the modulation of the expression of genes associated with cell proliferation, lipid and cholesterol synthesis, extracellular matrix components, intercellular interacting receptors and MK and platelet functions reflected the chronological development of the productive cells and pointed to unsuspected pathways. Surprisingly, SR1 only affected the gene expression profile of D10 CD34+CD41+CD9- cells; thus, as compared to these cells and those present on D14, the poorly productive D10 CD34+CD41+CD9- cells obtained in the absence of SR1 and the two unproductive populations present on D10 displayed an intermediate gene expression pattern. In other words, the ability to generate proplatelets between D10 and D14 appeared to be linked to the capacity of SR1 to delay MK differentiation, meanwhile avoiding intermediate and inappropriate genetic perturbations. Paradoxically, the D14 CD34+CD41+CD9- cells obtained under SR1- or SR1+ conditions were virtually identical, raising the question as to whether their strong differences in terms of proplatelet production, in the absence of SR1 and between D14 and D17, are mediated by miRNAs or by memory post-translational regulatory mechanisms.

cell biology↗