Skip to content

Latest commit

 

History

History
39 lines (31 loc) · 1.5 KB

README.md

File metadata and controls

39 lines (31 loc) · 1.5 KB
Raw

Synthesizability of generative methods

This repository contains codes and results to analyze synthesizability using ASKCOS for molecular generation and optimization algorithms.
A more detailed explanation of the project can be found in our paper.

Usage with ASKCOS API

This analysis need the API version of ASKCOS retrosynthesis tree builder. If you want to test a set of molecules in SMILES (e.g. test.csv), please replace the HOST address in tb_analysis/batch_TB.py to your ASKCOS server IP, then

python batch_TB.py -i test.csv

The results will be stored at the same directory with python file in json format. One can also define the index of molecule to start with (default is 0) and the column name for SMILES strings (default is ""SMILES"). To see help message:

python batch_TB.py -h

Usage with local ASKCOS

Under development

Dependencies

If you just want to test synthesizability

  • Python (3.6.9)
  • numpy (1.16.4)
  • pandas (0.24.2)
  • requests (2.22.0)

If you want to use the generative models and benchmarks to reproduce the result

Please refer to MOSES for dependencies of distribution learning implementations and Guacamol Baselines for dependencies of goal-directed learning implementations. The benchmarks objective functions can be found in goal_directed_generation/guacamol_local