SABLE: AI-Driven Framework for Multi-Objective Drug Optimization
A new open-source framework named SABLE (Synthetically-accessible Agentic Bayesian Ligand Exploration) has been developed by researchers to enhance chemical structure optimization in hit-to-lead drug discovery through natural-language orchestration. Utilizing a large language model (LLM), it interprets user-defined objectives and directs tasks, while specialized tools carry out reaction-templated analog enumeration, predict physicochemical and ADMET properties, score based on structure-affinity, and perform Bayesian optimization. This computational workflow mirrors the analytical and prioritization phases of the design-make-test-analyze cycle, ensuring provenance for each numerical result. In both single- and multi-objective optimization studies, SABLE effectively enriches candidate sets while only evaluating a portion of the chemical space. The framework tackles the complexities of balancing potency, selectivity, physicochemical, pharmacokinetic, safety, and synthetic constraints. The paper can be found on arXiv with the identifier 2608.11483.
Key facts
- SABLE is an open-source framework for hit-to-lead optimization.
- It uses natural-language orchestration to guide chemical structure optimization.
- An LLM interprets user-defined goals and routes tasks.
- Specialized tools include reaction-templated analog enumeration, property prediction, affinity scoring, and Bayesian optimization.
- The workflow provides provenance for each numerical output.
- It is a computational twin of the design-make-test-analyze cycle.
- SABLE enriches candidate sets in single- and multi-objective studies.
- The paper is available on arXiv (2608.11483).
Entities
Institutions
- arXiv