(645a) Hybrid Synthetic Chemistry - Synthetic Biology Reaction Network for Reaction Routes Optimization
AIChE Annual Meeting
2021
2021 Annual Meeting
Catalysis and Reaction Engineering Division
Reaction Path Analysis (Virtual)
Wednesday, November 17, 2021 - 4:20pm to 4:45pm
In this work, 21 million organic reactions from Reaxys database and all available (60 thousand) metabolic reactions from Kyoto Encyclopedia of Genes and Genomes (KEGG) database were digitalized and merged to generate a hybrid reaction network. The objective was to quantify the added values of the sparse metabolic dataset onto the organic dataset, for the guidance of retrosynthetic planning from target molecules to small precursor molecules. To do this, we assigned metrics for early-stage assessment of synthetic routes, including evaluating the carbon efficiency, capital costs of building blocks and length of synthetic steps. A reinforcement learning approach was employed to learn from the bulk simulated experience of synthetic routes generated from the organic, metabolic and hybrid networks. A reinforcement learning policy model predicted the expected values of all reactions in the chemical space, and given a target molecule, the model would suggest (near)-optimal multi-step reaction choices for retrosynthetic planning. The comparison among the three networks demonstrates the added values of metabolic reactions to improve redox efficiency, ease reaction pressure/temperature and enable synthetic shortcuts.
We also proposed a (near)-optimal synthetic route of a pharmaceutical, taxol, suggested by the policy model from the hybrid network. Organic synthesis of taxol was believed to be a challenge from previous CASP tools. Here we conducted later-stage assessment for the biochemical synthetic routes of taxol, including evaluating its ease of separation, exergetic efficiencies, sustainability and etc., to demonstrate the feasibility of the proposed synthetic route.