Specifying Chemprop Version

This release specifies that Chemprop should be version 1.6.1 in the setup.py for pip installation since the new version of Chemprop (2.0) currently breaks Chemprop-related imports in SyntheMol.

NMI paper in README

Added information about our Nature Machine Intelligence paper in the README.

Clean building blocks

Cleaned up unnecessary columns from the REAL building blocks file to reduce the file size.

Release for Zenodo DOI

Release to trigger a Zenodo DOI creation

Public release

Public release of SyntheMol GitHub repo and PyPI package.

Refactor and reproducibility check

Completed a major refactor of the SyntheMol code to improve ease of use (i.e., as pip package with synthemol command) and verified reproducibility of prior results.

Complete analysis prior to cleanup

This release marks the state of the code after completing analysis of the generated, synthesized, and experimentally validated molecules and writing the paper but prior to cleaning up the repo (e.g., changing file names, removing unused scripts).

First round of generations for experimental testing

This release marks the state of the code used during the first round of generation for experimental testing. Molecules were generated using 20k rollouts for RF RDKit, chemprop, chemprop RDKit, and random models. Each was filtered using train hits Tversky similarity <= 0.5, ChEMBL antibiotics Tversky similarity <= 0.5, top 20% scores, and k-means clustering into 50 clusters. Among these 200 molecules, 103 are available from Enamine and will be purchased. All instructions for reproducing these results are in the README.