Improved decision making with similarity based machine learning: Applications in chemistry

TL;DR

This paper introduces similarity-based machine learning (SML) for chemistry, enabling effective decision making in data-scarce scenarios by selecting relevant training data dynamically, thus reducing data requirements.

Contribution

The paper presents a novel SML approach that adapts training data selection on-the-fly for specific queries, improving performance in chemical applications with limited data.

Findings

SML achieves competitive performance with less data

Application to quantum chemistry and synthesis planning demonstrates effectiveness

Derived relationship between feature space properties and model accuracy

Abstract

Despite the fundamental progress in autonomous molecular and materials discovery, data scarcity throughout chemical compound space still severely hampers the use of modern ready-made machine learning models as they rely heavily on the paradigm, 'the bigger the data the better'. Presenting similarity based machine learning (SML), we show an approach to select data and train a model on-the-fly for specific queries, enabling decision making in data scarce scenarios in chemistry. By solely relying on query and training data proximity to choose training points, only a fraction of data is necessary to converge to competitive performance. After introducing SML for the harmonic oscillator and the Rosenbrock function, we describe applications to scarce data scenarios in chemistry which include quantum mechanics based molecular design and organic synthesis planning. Finally, we derive a relationship between the intrinsic dimensionality and volume of feature space, governing the overall model accuracy.