2021.01 Public Release

askcos-core 2021.01

User notes:

• Add functions for reordering precursors using SCScore
• Add SCScore termination criteria to tree builder
• Enable disabling of price lookup in tree builder
• Filter templates by attribute filter
• Add solvent scores to the condition recommender
• Add new version of MCTS tree builder
• Add pathway ranking and clustering model
• Add QM descriptor model
• Add QM-based universal regioselectivity model
• Add new quantitative reaction condition recommendation model

Developer notes:

• Update Docker image Python version to 3.7 and RDKit version to 2020.03.06
• New pathway enumeration algorithm for tree builder
• Improve flexibility of mongodb connection parameters
• Return template rank with one-step predictions
• Switch to C++ version of rdchiral for improved performance
• Return graph attributes, including tree depth, in treedata json
• Add conda environment file to facilitate local installation of dependencies
• Adjust impurity predictor inspector to use forward predictor score

Bug fixes:

• Fix template drawing for aromatic structures
• Improved tracking of different buyable sources
• Tree builder no longer returns cyclic pathways
• Pass template_set to retro transformer
• Pass template_set to retrieve_template_metadata
• Do not run pathway ranking when there is 1 tree

Breaking changes:

• Return values changed for neural net context recommender get_n_conditions method
• Return values changed for MCTS tree builder tree_status method
• Return values changed for MCTS tree builder get_buyable_paths method

2020.07 Public Release

askcos-core 2020.07

Release Notes

The following release notes describe changes since ASKCOS v0.4.1.

Developer notes:

• Refactored MCTS code to decouple pure python code from celery infrastructure
• Added Makefile to facilitate building docker images
• Renamed makeit module to askcos
• Revert to default coordinate generator for RDKit drawing
• Use askcos-base as base image for Docker build

Bug fixes:

• Running the main tree.builder.py file as a script or the tree builder unit test with multiprocessing did not work
• Atomic identity should not change in a tree builder reaction prediction
• Catch general selectivity error when target is not recovered
• Tree builder returns the incorrect supporting template
• Fix template db lookup in tree builder to include template_set
• Catch error when retro template cant be parsed by rdchiral when reversed
• Do not use isomeric smiles for TFFP prediction