Data-driven reactivity prediction of targeted covalent inhibitors using computed quantum features for drug discovery

TL;DR

This paper introduces a quantum feature-based data-driven approach to predict the reactivity of covalent inhibitors, demonstrating improved accuracy and potential for quantum computing integration in drug discovery.

Contribution

It presents a novel quantum embedding and simulation workflow to generate molecular features for reactivity prediction, advancing the integration of quantum computing in drug discovery.

Findings

Quantum fingerprinting effectively clusters molecules by warhead properties.

Predictions improve with larger active spaces and longer evolution times.

The approach is compatible with current and future quantum hardware.

Abstract

We present an approach to combine novel molecular features with experimental data within a data-driven pipeline. The method is applied to the challenge of predicting the reactivity of a series of sulfonyl fluoride molecular fragments used for drug discovery of targeted covalent inhibitors. We demonstrate utility in predicting reactivity using features extracted from a workflow which employs quantum embedding of the reactive warhead using density matrix embedding theory, followed by Hamiltonian simulation of the resulting fragment model from an initial reference state. These predictions are found to improve when studying both larger active spaces and longer evolution times. The calculated features form a `quantum fingerprint' which allows molecules to be clustered with regard to warhead properties. We identify that the quantum fingerprint is well suited to scalable calculation on future quantum computing hardware, and explore approaches to capture results on current quantum hardware using error mitigation and suppression techniques. We further discuss how this general framework may be applied to a wider range of challenges where the potential for future quantum utility exists.