ASKCOS: An open-source, microservice-based web app for automatic and interactive synthesis planning

Synthesis planning is a fundamental problem in organic chemistry. Given a target molecule to be synthesized, chemists design routes by breaking it down to smaller and smaller intermediates, and eventually to building blocks which are commercially or synthetically accessible. This interactive process has benefited tremendously from advancement in machine learning (ML) in the past few years. Template-based and template-free ML models for one-step retrosynthesis have been developed, as well as various search strategies for multi-step planning. These models and algorithms have been integrated into open-source tools such as ASKCOS for computer-aided synthesis planning (CASP), which has facilitated the workflow of many chemists since its initial release.

Here we present the latest version of ASKCOS (available at askcos.mit.edu), a fully refactored and more powerful version of ASKCOS with a microservice-based architecture. ASKCOS brings together several machine learning models and cheminformatics tools for synthesis planning, including and beyond retrosynthetic analysis. The monolithic Python-based backend has been re-modularized into individual containerized services, which ensures compatibility with any existing workflow or codebase of the user. ASKCOS is easy to use with a user-friendly web interface or API, and easy to deploy with few commands. We hope that ASKCOS can be a potent tool for many organic chemists, and would like to call for open-source contributions to help advance the CASP field.