Chimera: Accurate retrosynthesis prediction by ensembling models with diverse inductive biases

2024-12-06

arXiv.org

AI Summary

Chemical synthesis remains a critical bottleneck in small-molecule discovery and manufacturing. Current AI-driven retrosynthetic models exhibit poor generalization to rare yet chemically essential reactions and suffer from hallucinatory errors, undermining search algorithm reliability and chemists’ trust. To address these limitations, we propose Chimera—the first learning-based ensemble framework explicitly designed for chemical reasoning diversity. Chimera integrates two state-of-the-art architectures—Transformer and graph neural network—via learnable weighted ensembling and distributionally robust calibration, thereby mitigating the inductive biases inherent in individual models. Extensive evaluation across multiple standardized benchmarks demonstrates significant performance gains over leading methods. Expert validation by organic chemistry PhDs confirms superior synthetic plan quality and interpretability. Furthermore, Chimera achieves exceptional out-of-distribution generalization on proprietary pharmaceutical industry datasets, underscoring its practical utility in real-world drug discovery pipelines.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Planning and conducting chemical syntheses remains a major bottleneck in the discovery of functional small molecules, and prevents fully leveraging generative AI for molecular inverse design. While early work has shown that ML-based retrosynthesis models can predict reasonable routes, their low accuracy for less frequent, yet important reactions has been pointed out. As multi-step search algorithms are limited to reactions suggested by the underlying model, the applicability of those tools is inherently constrained by the accuracy of retrosynthesis prediction. Inspired by how chemists use different strategies to ideate reactions, we propose Chimera: a framework for building highly accurate reaction models that combine predictions from diverse sources with complementary inductive biases using a learning-based ensembling strategy. We instantiate the framework with two newly developed models, which already by themselves achieve state of the art in their categories. Through experiments across several orders of magnitude in data scale and time-splits, we show Chimera outperforms all major models by a large margin, owing both to the good individual performance of its constituents, but also to the scalability of our ensembling strategy. Moreover, we find that PhD-level organic chemists prefer predictions from Chimera over baselines in terms of quality. Finally, we transfer the largest-scale checkpoint to an internal dataset from a major pharmaceutical company, showing robust generalization under distribution shift. With the new dimension that our framework unlocks, we anticipate further acceleration in the development of even more accurate models.

Problem

• AI models struggle with rare but critical chemical reactions
• Existing models produce incorrect predictions misaligned with chemists' expectations
• Multi-step synthesis algorithms fail due to unreliable model outputs

Innovation

• Ensembling diverse inductive bias models
• Learning-based fusion of multiple predictions
• Zero-shot transfer to industrial datasets