v1.5.0

This new release fixes some critical bugs, improves the default design settings, and adds new options for complex reprediction for hard targets:

• Added option to predict MPNN-optimised designs using initial guess (Bennett et al. 2023) or big bang (Schweke et al. 2023, for large complexes). These options improve in silico design success for difficult targets. By default we still recommend to use single sequence mode (default option).
• Changed 'inter_contact_number' from 1 to 2 by default, which strongly reduces the number of trajectories that ignore selected hotspots for binding. In other words, most binder designs should now bind to the hotspot residues you specify in the settings.
• Fixed memory leakage
• Fixed improper saving of trajectory metrics into csv

v1.2.0

• fixed contact filter for peptides, which was causing very short peptide binders to be rejected
• avoid the saving of duplicate sequences during MPNN redesign, which is very unlikely but can happen when designing very short peptides, where multiple trajectories converge to the same sequence
• added extensive checks for the installation script to make sure each step is completed fully before proceeding
• removed default anaconda channel dependency
• added libgfortran5 to installation requirements
• added live trajectory and accepted design counters to the colab notebook
• fixed hydrophobicity calculation for binder surface, there was a bug where the surface taken into account was from the whole complex instead of just the binder alone
• colab target settings are now saved in the design output folder on Google drive and can be reloaded for continuing the design campaign
• added options into settings_advanced jsons to manually set AF2 params directory, or dssp path or dalphaball path. If left empty, it will set the default installation paths
• added more relaxed filter settings for normal proteins and peptides
• added more advanced setting files allowing to redesign interface with MPNN, as well as increased flexibility of the target by masking the template sequence during design and reprediction
• fixed mpnn sequence generation where batch size did not correspond to number of generated sequences

v1.1.0

New update fixes problems with installation and compatibility:

• updated to python 3.10
• updated biopython to >1.8
• added CUDA version dependence for the installation script to choose correct jaxlib version
• merged all package requirement into a single conda command
• updated Readme
• fixed relative paths for running the scripts to enable execution from anywhere

v1.0.0

First public release of BindCraft