Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 24 min ago
Peer-reviewedCC BYNature Computational Science

SmileyLlama: modifying large language models for directed chemical space exploration

  • Joseph M. Cavanagh1
  • Kunyang Sun1
  • Andrew Gritsevskiy2
  • Dorian Bagni1
  • Yingze Wang1
  • Thomas D. Bannister3
  • Teresa L. Head-Gordon4,5,6
6 affiliations
  1. 1Kenneth S. Pitzer Theory Center and Department of Chemistry, University of California, Berkeley, Berkeley, CA, USA
  2. 2Promontory Labs, San Francisco, CA, USA
  3. 3Department of Molecular Medicine, The Herbert Wertheim UF Scripps Institute for Biomedical Innovation and Technology, Jupiter, FL, USA
  4. 4Chemical Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA, USA. [email protected]
  5. 5Departments of Bioengineering and Chemical and Biomolecular Engineering, University of California, Berkeley, Berkeley, CA, USA. [email protected]
  6. 6Kenneth S. Pitzer Theory Center and Department of Chemistry, University of California, Berkeley, Berkeley, CA, USA. [email protected]
Open at sourcePDFDOICite

Abstract

Here we show that large language models (LLMs) can be transformed via supervised fine-tuning of engineered prompts into SmileyLlama for exploring the chemical space of drug molecules. We benchmark SmileyLlama against pretrained LLMs and chemical language models trained from scratch for generating valid and novel drug-like molecules, and use direct preference optimization to both improve SmileyLlama's adherence to a prompt and as part of the iMiner reinforcement learning framework to predict molecules with optimized three-dimensional conformations and high binding affinity to drug targets. By training an LLM to speak directly as a chemical language model, while retaining most of its natural language capabilities, we show that SmileyLlama can reliably generate molecules with user-specified properties rather than acting only as a chatbot with knowledge of chemistry or as a virtual assistant. While SmileyLlama is geared toward drug discovery, the supervised fine-tuning/direct preference optimization/LLM framework can be extended to other chemical, biological and materials applications.

Related

  1. Syndirella: synthesis-directed fragment elaboration enables extensive and cost-effective binding site exploration beyond catalog compounds
    Kate K. Fieseler, Max Winokan, Joseph A. Morrone +3
    Chemical sciencePeer-reviewedCC BYFull textOrganic chemistryEurope PMCPMC
  2. Bridging the Literature-to-Laboratory Gap with Language Models and Active Learning for Reaction Feasibility Prediction and Autonomous Exploration
    Haowen Zhong, Yi Yang, Jiekai Jia +7
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
  3. QALPA: Property-guided diffusion modeling for efficient exploration of chemical spaces of flexible molecules
    Michael Hanna, Julian Cremer, Zekiye Erarslan +1
    arXivPreprintCC BY-NC-ND 4.0Full textPhysical chemistryarXiv
  4. Comprehensive Chemical Space Exploration in De Novo Drug Design: An Algorithmically Constrained, Parallel Dual-Paradigm Framework Validated on BACE1
    Thang T. Dang, Mai N. Luong
    ChemRxivPreprintCC BYGeneral chemistryChemRxiv
  5. Measuring chemical safety and refusal in language models
    Ashutosh S. Jogalekar
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv

Cite

BibTeXRIS
BibTeX
@article{cavanagh2026smileyllama,
  title = {{SmileyLlama: modifying large language models for directed chemical space exploration}},
  author = {Joseph M. Cavanagh and Kunyang Sun and Andrew Gritsevskiy and Dorian Bagni and Yingze Wang and Thomas D. Bannister and Teresa L. Head-Gordon},
  journal = {Nature Computational Science},
  year = {2026},
  month = may,
  doi = {10.1038/s43588-026-00986-y},
  url = {https://doi.org/10.1038/s43588-026-00986-y},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - SmileyLlama: modifying large language models for directed chemical space exploration
AU  - Joseph M. Cavanagh
AU  - Kunyang Sun
AU  - Andrew Gritsevskiy
AU  - Dorian Bagni
AU  - Yingze Wang
AU  - Thomas D. Bannister
AU  - Teresa L. Head-Gordon
PY  - 2026
DA  - 2026/05/11
JO  - Nature Computational Science
DO  - 10.1038/s43588-026-00986-y
UR  - https://doi.org/10.1038/s43588-026-00986-y
AB  - Here we show that large language models (LLMs) can be transformed via supervised fine-tuning of engineered prompts into SmileyLlama for exploring the chemical space of drug molecules. We benchmark SmileyLlama against pretrained LLMs and chemical language models trained from scratch for generating valid and novel drug-like molecules, and use direct preference optimization to both improve SmileyLlama's adherence to a prompt and as part of the iMiner reinforcement learning framework to predict molecules with optimized three-dimensional conformations and high binding affinity to drug targets. By training an LLM to speak directly as a chemical language model, while retaining most of its natural language capabilities, we show that SmileyLlama can reliably generate molecules with user-specified properties rather than acting only as a chatbot with knowledge of chemistry or as a virtual assistant. While SmileyLlama is geared toward drug discovery, the supervised fine-tuning/direct preference optimization/LLM framework can be extended to other chemical, biological and materials applications.
SN  - 2662-8457
ER  -