Evaluating Molecule Synthesizability via Retrosynthetic Planning and Reaction Prediction

2024-11-13

AI Summary

In drug design, AI-generated molecules often exhibit a trade-off between high predicted bioactivity and synthetic tractability—either highly potent but synthetically challenging, or easily synthesized but pharmacologically weak. Conventional synthetic accessibility scores (e.g., SA Score, RAscore) rely on heuristic or statistical proxies without verification against actual synthetic routes. To address this, we propose the Round-trip Score, a novel metric grounded in bidirectional consistency between retrosynthetic planning and forward reaction prediction. It integrates state-of-the-art, large-data-trained retrosynthesis models (e.g., G2G, Molecular Transformer) with high-accuracy forward reaction predictors to form an end-to-end feasibility validation framework. Evaluated across multiple leading molecular generative models, the Round-trip Score demonstrates strong correlation with experimentally observed synthesis success rates—significantly outperforming traditional metrics—and enables reliable, interpretable prioritization of AI-designed compounds. This work establishes the first paradigm shift from empirical scoring to path-verified synthetic assessability.

Technology Category

• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

A significant challenge in wet lab experiments with current drug design generative models is the trade-off between pharmacological properties and synthesizability. Molecules predicted to have highly desirable properties are often difficult to synthesize, while those that are easily synthesizable tend to exhibit less favorable properties. As a result, evaluating the synthesizability of molecules in general drug design scenarios remains a significant challenge in the field of drug discovery. The commonly used synthetic accessibility (SA) score aims to evaluate the ease of synthesizing generated molecules, but it falls short of guaranteeing that synthetic routes can actually be found. Inspired by recent advances in top-down synthetic route generation and forward reaction prediction, we propose a new, data-driven metric to evaluate molecule synthesizability. This novel metric leverages the synergistic duality between retrosynthetic planners and reaction predictors, both of which are trained on extensive reaction datasets. To demonstrate the efficacy of our metric, we conduct a comprehensive evaluation of round-trip scores across a range of representative molecule generative models.

Problem

• Balancing pharmacological properties and synthesizability in drug design
• Improving synthetic accessibility score accuracy for molecule evaluation
• Developing data-driven metric using retrosynthetic and reaction prediction

Innovation

• Retrosynthetic planners for molecule synthesizability evaluation
• Reaction predictors trained on extensive datasets
• Data-driven metric combining retrosynthesis and reaction prediction