VALID-Mol: a Systematic Framework for Validated LLM-Assisted Molecular Design

2025-06-29

AI Summary

Large language models (LLMs) exhibit poor chemical validity and low efficiency (only 3%) in molecular design, generating numerous chemically invalid or unsynthesizable structures. Method: This work proposes a scientific-constraint-guided LLM framework for molecular design, integrating domain-adapted fine-tuned models, chemistry-aware prompt engineering, and automated structural validity verification to ensure generated molecules satisfy chemical validity, synthetic accessibility, and target property optimization. Contribution/Results: The approach elevates the rate of chemically valid molecule generation to 83% and achieves up to a 17-fold improvement in computationally predicted target protein binding affinity. To our knowledge, this is the first systematic, generalizable, and reproducible interdisciplinary generative paradigm—termed “chemical-rule embedding → LLM generation → closed-loop validation”—establishing a methodological foundation and practical blueprint for AI-driven drug discovery.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Large Language Models (LLMs) demonstrate remarkable potential for scientific discovery, but their application in domains requiring factual accuracy and domain-specific constraints remains challenging. In molecular design for drug discovery, LLMs can suggest creative molecular modifications but often produce chemically invalid or impractical structures. We present VALID-Mol, a systematic framework for integrating chemical validation with LLM-driven molecular design that increases the rate of generating valid chemical structures from 3% to 83%. Our approach combines methodical prompt engineering, automated chemical validation, and a fine-tuned domain-adapted LLM to ensure reliable generation of synthesizable molecules with improved properties. Beyond the specific implementation, we contribute a generalizable methodology for scientifically-constrained LLM applications, with quantifiable reliability improvements. Computational predictions suggest our framework can generate promising candidates for synthesis with up to 17-fold computationally predicted improvements in target affinity while maintaining synthetic accessibility. We provide a detailed analysis of our prompt engineering process, validation architecture, and fine-tuning approach, offering a reproducible blueprint for applying LLMs to other scientific domains where domain-specific validation is essential.

Problem

• Ensures chemically valid molecular structures in LLM-assisted drug design
• Improves valid chemical structure generation rate from 3% to 83%
• Provides a generalizable framework for domain-constrained LLM applications

Innovation

• Methodical prompt engineering for reliable outputs
• Automated chemical validation ensuring synthesizable molecules
• Fine-tuned domain-adapted LLM for scientific constraints