SABLE Framework Optimizes Drug Discovery with AI Agents

Kelvin P. Idanwekhai, Enes Kelestemur, Benjamin Strickland, Matthew Hart, Steini Davidsson, Angelos Angelopoulos, Ron Alterovitz, Marcello DeLuca, Alexander Tropsha· August 13, 2026 View original

Key takeaways

  • SABLE uses LLMs and specialized tools for automated, multi-objective drug candidate optimization.
  • The framework acts as a computational twin for early-stage drug discovery, enhancing efficiency.
  • Its modular design allows for flexible integration with various chemical characterization backends.
  • SABLE helps prioritize synthetically constrained analogs, accelerating the hit-to-lead process.

Who benefits

PharmaceuticalsBiotechnologyChemicalsLife Sciences

Summary

SABLE is an open-source, agentic framework that uses natural language orchestration to guide chemical structure optimization in drug discovery, integrating LLMs with specialized tools for iterative design. It acts as a computational twin for the analytical and prioritization stages of the design-make-test-analyze cycle.

This research introduces SABLE, an open-source framework designed to streamline the complex process of hit-to-lead optimization in drug discovery. SABLE leverages large language models for natural language interpretation of user goals, orchestrating various specialized tools. These tools handle tasks such as enumerating chemical analogs, predicting properties like ADMET, and scoring molecular affinity. The framework effectively mimics the analytical and prioritization phases of the traditional drug discovery cycle, providing clear provenance for all computational outputs. Its modular design allows for easy replacement of tools and characterization backends, making it highly adaptable. SABLE has demonstrated its ability to enrich candidate sets for defined computational objectives while efficiently exploring the vast chemical search space.

Why it matters

Professionals in pharmaceutical R&D can leverage this framework to accelerate early-stage drug discovery by automating and optimizing the design of new drug candidates, potentially reducing costs and time-to-market.

How to implement this in your domain

  1. 1Explore SABLE's open-source code to understand its architecture and integration points.
  2. 2Integrate SABLE with existing in-house computational chemistry tools and databases.
  3. 3Define specific multi-objective optimization goals for drug candidates using natural language prompts.
  4. 4Utilize the framework to generate and prioritize synthetically accessible drug analogs.
  5. 5Evaluate the provenance of numerical outputs to ensure transparency and reproducibility in drug design.

Original post by Kelvin P. Idanwekhai, Enes Kelestemur, Benjamin Strickland, Matthew Hart, Steini Davidsson, Angelos Angelopoulos, Ron Alterovitz, Marcello DeLuca, Alexander Tropsha

"arXiv:2608.11483v1 Announce Type: new Abstract: Hit-to-lead optimization requires iterative design of hit analogs across competing potency, selectivity, physicochemical, pharmacokinetic, safety, and synthetic constraints. We present SABLE (Synthetically-accessible Agentic Bayesia…"

View on X

Originally posted by Kelvin P. Idanwekhai, Enes Kelestemur, Benjamin Strickland, Matthew Hart, Steini Davidsson, Angelos Angelopoulos, Ron Alterovitz, Marcello DeLuca, Alexander Tropsha on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research