Integrating Synthetic Accessibility with AI-based Generative Drug Design

Abstract

Generative models are frequently used for de novo design in drug discovery projects to propose new molecules. However, the question of whether or not the generated molecules can be synthesized is not systematically taken into account during generation, even though being able to synthesize the generated molecules is a fundamental requirement for such methods to be useful in practice. Methods have been developed to estimate molecule synthesizability, but, so far, there is no consensus on whether or not a molecule is synthesizable. In this paper we introduce the Retro-Score (RScore), which computes a synthetic feasibility score of molecules by performing a full retrosynthetic analysis through our data-driven synthetic planning software Spaya, and its dedicated API: Spaya-API (https://spaya.ai). After a comparison of RScore with other synthetic scores from the literature, we describe a pipeline to generate molecules that validate a list of targets while still being easy to synthesize. We further this idea by performing experiments comparing molecular generator outputs across a range of constraints and conditions. We show that the RScore can be learned by a Neural Network, which leads to a new score: RSPred. We demonstrate that using the RScore or RSPred as a constraint during molecular generation enables to obtain more synthesizable solutions, with higher diversity. The open-source Python code containing all the scores and the experiments can be found on https://github.com/iktos/generation-under-synthetic- constraint.

Navigating with chemometrics and machine learning in chemistry

“…It allows both retro- and forward- prediction of reactions. It avoids predicting patented synthetic routes (Klucznik et al 2018 ) Spaya AI Based on deep machine learning using databases such as Mcule, Chemspace, EMolecules (Parrot et al 2021 ) LillyMol Utilizes machine learning approach, atom mapping and train reaction transformation rules (Watson et al 2019 ) AutoSynRoute Utilizes Montel-Carlo tree search with heuristic scoring function, transformer-type-seq-2-seq model (Lin et al 2020 ) AiZynthFinder Monte Carlo tree search by ANN policy that allowed prioritizing reaction templates to generate novel precursors (Genheden et al 2020 ) IBM RXN for Chemistry Forward prediction, one-step retrosynthetic tool that uses seq-2-seq database, natural language approach and trained on automatic extracted chemical data (IBM 2018 ) ICSYNTH Utilizes machine learning to generate chemical rules from SPRESI database (Bøgevig et al 2015 ) PostEra Manifold Open-source retrosynthesis tool that allows search for different synthetic routes and generates comparison of raw materials from different vendors (PostEra 2021 ) …”