Agent-driven Model Development for RNA 3D Structure Prediction
- 1Sanofi
Abstract
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.
Related
- LifeLong Digital Twin: A Unified Modeling Paradigm and Agent Harness for Event-Driven Lifelong Health State Trajectories
- Scale-free Niche Construction: expanding agent-microenvironment co-development to unconventional substrates
- A neuroevolution-driven agent-based model of coral larvae settlement
- A Shared-Task-State-Driven Multi-Agent Framework for Knowledge Graph Construction in Substation Operation and Maintenance
- CausalOdyssey: Multi-Agent Local LLM Reasoning with Per-Patient Causal Structure for Rare Disease Retrieval
BibTeX
@misc{majewski2026agentdriven,
title = {{Agent-driven Model Development for RNA 3D Structure Prediction}},
author = {M. Majewski and L. Malo and A. Montero-Blay and M. Marengo and P. Gkeka and H. Minoux},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.08.749228},
url = {https://www.biorxiv.org/content/10.64898/2026.09.08.749228v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Agent-driven Model Development for RNA 3D Structure Prediction
AU - M. Majewski
AU - L. Malo
AU - A. Montero-Blay
AU - M. Marengo
AU - P. Gkeka
AU - H. Minoux
PY - 2026
DA - 2026/09/08
PB - bioRxiv
DO - 10.64898/2026.09.08.749228
UR - https://www.biorxiv.org/content/10.64898/2026.09.08.749228v1
AB - 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.
ER -