Data‐driven approaches for identifying hyperparameters in multi‐step retrosynthesis

Abstract

Multi‐step retrosynthesis problem can be solved by a search algorithm, such as Monte Carlo tree search (MCTS). The performance of multistep retrosynthesis, as measured by a trade‐off in search time and route solvability, therefore depends on the hyperparameters of the search algorithm. In this paper, we demonstrated the effect of three MCTS hyperparameters (number of iterations, tree depth, and tree width) on metrics such as Linear integrated speed‐accuracy score (LISAS) and Inverse efficiency score which consider both route solvability and search time. This exploration was conducted by employing three data‐driven approaches, namely a systematic grid search, Bayesian optimization over an ensemble of molecules to obtain static MCTS hyperparameters, and a machine learning approach to dynamically predict optimal MCTS hyperparameters given an input target molecule. With the obtained results on the internal dataset, we demonstrated that it is possible to identify a hyperparameter sets which outperform the current AiZynthFinder default setting and appeared optimal across a variety of target input molecules, both on proprietary and public datasets. The settings identified with the in‐house dataset reached a solvability of 93% and median search time of 151s for the in‐house dataset, and a 74% solvability and 114s for the ChEMBL dataset. These numbers can be compared to the current default settings which solved 85% and 73% during a median time of 110s and 84s, for in‐house and ChEMBL, respectively.

AiZynthFinder 4.0: developments based on learnings from 3 years of industrial application

“…For ChEMBL, we only find solutions to 71% of the targets, somewhat lower than the AstraZeneca sets. However, this could be explained by the extended number of iterations used for the AstraZeneca sets as recommended in a recent study of search hyperparameters [ 61 ]. For the GDB set we only find solutions to about 10%, highlighting a disconnection of the current template-based model trained on historical reaction data, with the chemistry needed to find synthesis routes for the enumerated, and therefore potentially non-synthesizable, GDB compounds.…”

Do Chemformers Dream of Organic Matter? Evaluating a Transformer Model for Multistep Retrosynthesis

“…As such, the number of iterations is not a limiting factor for the template-based model. The hyperparameters for the MCTS search, including the number of iterations, were set to their default values since our intention was to evaluate the Chemformer within the production framework . For the Chemformer predictions, we used a beam size of 10, whereas for the template-based model, we added the top-50 predictions to the search tree.…”

“…The maximum search time was set to 300 s for both the template-based model and Chemformer. According to our recent paper on hyperparameter-tuning MCTS, there is only a very minor difference in solvability when setting the number of iterations to 100 compared to 400, while the search time is often beyond 300 s for 400 iterations. As such, the number of iterations is not a limiting factor for the template-based model.…”

Artificial Intelligence Methods and Models for Retro-Biosynthesis: A Scoping Review

“…Furthermore, the configuration of the model significantly influences the success of route discovery. 99 Template reactions model a finite number of transformations, which may lack exhaustiveness. As an alternative to template-based formalism, template-free singlestep approaches have also been employed in MCTS.…”

“…The data source for building templates is an important factor, as highlighted by Thakkar et al, wherein the impact of four data sets (AiZynthFinder, Pistachio, Reaxys, and USPTO) on MCTS performance is investigated. Furthermore, the configuration of the model significantly influences the success of route discovery . Template reactions model a finite number of transformations, which may lack exhaustiveness.…”