analiza aiqbench
Bolek is a 4B-parameter model that injects a Morgan fingerprint into an instruction-tuned LLM, reporting gains over its Qwen3-4B base (mean ROC/PR AUC 0.55 to 0.76) and over the 9B TxGemma-9B-Chat on most of 15 TDC tasks. The grounding claim rests on agreement with RDKit descriptors, but the training uses synthetic chain-of-thought, and on unseen endpoints it only matches TxGemma on five, so the generalisation case is weaker than the headline numbers suggest.
Bolek: A Multimodal Language Model for Molecular Reasoning
arXiv:2605.02745v2 Announce Type: replace Abstract: Molecular property models increasingly support high-stakes drug-discovery decisions, but their outputs are often difficult to audit: classical predictors return scores without rationale, while language models can produce fluent explanations weakly grounded in the input molecule. We introduce Bolek, a compact multimodal language model that grounds natural-language reasoning in molecular structure by injecting a Morgan fingerprint embedding into an instruction-tuned text decoder. Bolek is fine-tuned on molecular alignment tasks, including molecule description, RDKit descriptor prediction, and substructure detection, and on downstream reasoning over 15 TDC binary classification tasks using synthetic chains-of-thought anchored in concrete molecular features. Across these tasks, Bolek outperforms its Qwen3-4B-Instruct base on all endpoints in yes/no mode and on 13 of 15 in chain-of-thought mode, raising mean ROC/PR AUC from 0.55 to 0.76. It also outperforms TxGemma-9B-Chat on 13 of 15 binary classification tasks despite being less than half its size. Bolek's explanations are more grounded than those of the baseline LLMs: it cites numerical descriptors 10-100x more often per chain-of-thought, and the cited values agree strongly with RDKit for key descriptors such as TPSA, MolLogP, and MolWt (Spearman rho = 0.87-0.91). Generalisation extends beyond the training panel: on 15 unseen TDC classification endpoints, Bolek matches TxGemma on five, and it produces non-trivial rank correlations on three held-out regression endpoints despite never seeing downstream regression during training. These results suggest that targeted modality injection and reasoning supervision tied to verifiable molecular features can yield compact, auditable molecular reasoning models.
Abstract:Molecular property models increasingly support high-stakes drug-discovery decisions, but their outputs are often difficult to audit: classical predictors return scores without rationale, while language models can produce fluent explanations weakly grounded in the input molecule. We introduce Bolek, a compact multimodal language model that grounds natural-language reasoning in molecular structure by injecting a Morgan fingerprint embedding into an instruction-tuned text decoder. Bolek is fine-tuned on molecular alignment tasks, including molecule description, RDKit descriptor prediction, and substructure detection, and on downstream reasoning over 15 TDC binary classification tasks using synthetic chains-of-thought anchored in concrete molecular features. Across these tasks, Bolek outperforms its Qwen3-4B-Instruct base on all endpoints in yes/no mode and on 13 of 15 in chain-of-thought mode, raising mean ROC/PR AUC from 0.55 to 0.76. It also outperforms TxGemma-9B-Chat on 13 of 15 binary classification tasks despite being less than half its size. Bolek's explanations are more grounded than those of the baseline LLMs: it cites numerical descriptors 10-100x more often per chain-of-thought, and the cited values agree strongly with RDKit for key descriptors such as TPSA, MolLogP, and MolWt (Spearman rho = 0.87-0.91). Generalisation extends beyond the training panel: on 15 unseen TDC classification endpoints, Bolek matches TxGemma on five, and it produces non-trivial rank correlations on three held-out regression endpoints despite never seeing downstream regression during training. These results suggest that targeted modality injection and reasoning supervision tied to verifiable molecular features can yield compact, auditable molecular reasoning models.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Biomolecules (q-bio.BM)
arXiv:2605.02745 [cs.LG]
(or arXiv:2605.02745v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.02745 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 338 cuvinte
Subiecte conexe
SafeMol: Dual-Modality Safety Alignment for Molecular Multimodal Models
arXiv · 30 sept. 2026
Learning Molecular Representations from Cellular Phenotypes with Structure Preservation
arXiv · 5 aug. 2026
Hypothesis-and-Refinement Learning of Organic Structures from Multimodal Spectroscopic Data
arXiv · 23 iul. 2026
MM-Spectrum: Multimodal Multi-spectral Molecular Structural Elucidation with a Stable MoE Framework
arXiv · 28 aug. 2026