PharMistral: Drug generation for novel protein targets by Mistral Large Language Model
Computer Methods and Programs in Biomedicine Update, vol. 10 (2026), article 100257, Elsevier
Published online on , 1 citation on Google Scholar
In short
The paper presents PharMistral, a Mistral-7B model first trained on over 300,000 human protein sequences and over 1 million drug-like SMILES, then fine-tuned on about 300,000 protein-ligand pairs from ChEMBL to generate candidate drug molecules directly from a protein sequence. On unseen proteins, 99.5% of generated molecules were chemically valid, and 3,383 unique molecules passed drug-likeness and toxicity filters.
Cite
@article{Bendjeddou_2026,
title={PharMistral: Drug generation for novel protein targets by Mistral Large Language Model},
volume={10},
ISSN={2666-9900},
url={http://dx.doi.org/10.1016/j.cmpbup.2026.100257},
DOI={10.1016/j.cmpbup.2026.100257},
journal={Computer Methods and Programs in Biomedicine Update},
publisher={Elsevier BV},
author={Bendjeddou, Asma and Zeinalipour, Kamyar and Bardou, Dalal and Maggini, Marco and Scarselli, Franco and Bianchini, Monica},
year={2026},
month=Dec,
pages={100257}
}