MolRGen : A Training and Evaluation Setting for De Novo Molecular Generation with Reasonning Models

Philippe Formont, Maxime Darrin, Ismail Ben Ayed, Pablo Piantanida, [0.5ex] Université Paris-Saclay · 2026

Read MolRGen : A Training and Evaluation Setting for De Novo Molecular Generation with Reasonning Models by Philippe Formont, Maxime Darrin, Ismail Ben Ayed, Pablo Piantanida, [0.5ex] Université Paris-Saclay (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-18256.txt.

Description

Formont and co-authors introduce MolRGen, a training and test setting for de novo molecular generation with reasoning models. Rewards there can be computed without reference molecules, unlike past benchmarks. About 4,500 protein-pocket targets yield 50,000 prompts that join docking scores with drug-like traits. A diversity-aware top-k metric checks whether models make varied, high-scoring molecules. Fine-tuning a 128B model with GRPO raised scores but traded away diversity.

Catalog record

CREATORPhilippe Formont, Maxime Darrin, Ismail Ben Ayed, Pablo Piantanida, [0.5ex] Université Paris-Saclay
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE73786
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.18256

Browse the full library · About the Tower of Babel