Synergy Between Expert and Machine‐Learning Approaches Allows for Improved Retrosynthetic Planning

Abstract

When computers plan multistep syntheses, they can rely either on expert knowledge or information machine‐extracted from large reaction repositories. Both approaches suffer from imperfect functions evaluating reaction choices: expert functions are heuristics based on chemical intuition, whereas machine learning (ML) relies on neural networks (NNs) that can make meaningful predictions only about popular reaction types. This paper shows that expert and ML approaches can be synergistic—specifically, when NNs are trained on literature data matched onto high‐quality, expert‐coded reaction rules, they achieve higher synthetic accuracy than either of the methods alone and, importantly, can also handle rare/specialized reaction types.

Computer-generated “synthetic contingency” plans at times of logistics and supply problems: scenarios for hydroxychloroquine and remdesivir

“…In addition, the failure of the worldwide logistics and supply chains that has accompanied the COVID-19 pandemic might render some key substrates temporarily unavailable, in effect delaying the execution of the proven synthetic routes and calling for alternative synthetic solutions. Anticipating such complications, we harnessed the power of Chematica [9][10][11][12][13][14][15][16][17][18] an experimentally tested 10,11 platform for computer-assisted retrosynthesis of both known and unknown target molecules to design syntheses of HCQ and remdesivir. We were most interested in synthetic plans that would (1) commence from various inexpensive and popular starting materials (so that the syntheses minimize the abovementioned supply problems); (2) circumvent patented methodologies whenever possible; 16 and (3) minimize the use of expensive methodologies and/or reagents.…”

Data Augmentation and Pretraining for Template-Based Retrosynthetic Prediction in Computer-Aided Synthesis Planning

“…In contrast to the reaction prediction task in the forward direction, where a defined set of reaction conditions should lead to a single distribution of product molecules, a single-step retrosynthetic prediction takes the form of a one-to-many mapping, where the target could theoretically be made through a variety of different individual reaction steps. Retrosynthesis has seen increased attention from the data science and cheminformatics communities recently with a number of machine learning efforts leveraging reaction templates or rules, [1][2][3][4] techniques adapted from natural language processing, [5][6][7][8] and graph based models. 9,10 However, only the template and rulebased methods are capable of making a connection from the prediction directly back to the source of the template or rule, which is most likely a reaction that was successfully performed in a laboratory.…”

Computer-Assisted Planning of Hydroxychloroquine’s Syntheses Commencing from Inexpensive Substrates and Bypassing Patented Routes.

“…In addition, the failure of the worldwide logistics and supply chains that accompanies COVID-19 pandemic might render some key substrates temporarily unavailable, in effect delaying execution of the proven synthetic routes and calling for alternative synthetic solutions. Anticipating such complications, we harnessed the power of Chematica [8][9][10][11][12][13][14][15][16] -an experimentally-tested 9,10 platform for computer-assisted retrosynthesis of both known and unknown target molecules -to design syntheses of HCQ that would (1) commence from various inexpensive and popular starting materials (so that the syntheses minimize the abovementioned supply problems); (2) circumvent patented methodologies whenever possible 16 ;…”

“…Chematica is a sophisticated platform for fully automated design of pathways leading to arbitrary (i.e., both known and new) targets. The software combines elements of network theory 16,17 with an expert knowledge-base of synthetic transformations as well as multiple reaction-evaluation routines (based on machine learning, 11,12 quantum mechanics, 8,9 and molecular dynamics 9,13 ) to search over vast trees of synthetic possibilities. The reaction transforms (currently, ~ 100,000) are expert-coded based on the underlying reaction mechanisms and are broader than any specific literature precedents (for comparison with machine extraction of rules from reaction repositories, see 13 ).…”

“…Each rule specifies the scope of admissible substituents, accounts for stereo-and regiochemistry requirements, recognizes groups that must be protected under given reaction conditions, and identifies functionalities that are outright incompatible. The searches are guided by combinations of functions (either heuristic 8,9 or best-in-class AI-based 12 ) that score both synthetic positions as well as costs of individual reactions. The pathways identified by the program terminate in either commercially available chemicals (here, more than 200,000 molecules from Sigma-Aldrich catalogs, each with price per unit quantity; also see below for price re-scaling) or those already known in the literature (ca.…”