The Catechol Benchmark: Time-series Solvent Selection Data for Few-shot Machine Learning

TL;DR

This paper introduces the Catechol Benchmark, a novel time-series dataset for solvent selection in chemical processes, enabling machine learning models to predict yields across diverse continuous process conditions for sustainable manufacturing.

Contribution

The paper provides the first transient flow dataset for machine learning benchmarking in chemistry, focusing on solvent selection with continuous parameters, and evaluates various ML approaches on this challenging task.

Findings

Benchmarking results highlight the difficulty of modeling solvent selection.

Transfer learning improves prediction accuracy.

Active learning reduces data requirements for yield prediction.

Abstract

Machine learning has promised to change the landscape of laboratory chemistry, with impressive results in molecular property prediction and reaction retro-synthesis. However, chemical datasets are often inaccessible to the machine learning community as they tend to require cleaning, thorough understanding of the chemistry, or are simply not available. In this paper, we introduce a novel dataset for yield prediction, providing the first-ever transient flow dataset for machine learning benchmarking, covering over 1200 process conditions. While previous datasets focus on discrete parameters, our experimental set-up allow us to sample a large number of continuous process conditions, generating new challenges for machine learning models. We focus on solvent selection, a task that is particularly difficult to model theoretically and therefore ripe for machine learning applications. We showcase benchmarking for regression algorithms, transfer-learning approaches, feature engineering, and active learning, with important applications towards solvent replacement and sustainable manufacturing.