BECraft (BindEnergyCraft) is a drop-in extension of the protein–binder design pipeline first introduced in BindCraft (Pacesa et al., 2024). By adding an energy-based loss to the original AlphaFold2-back-propagation workflow, we improve binder designs while maintaining the user-friendly pipeline of BindCraft!
After cloning this repository, navigate to the becraft folder and run the BindCraft installation script:
bash install_bindcraft.sh --cuda '12.4' --pkg_manager 'conda'
Like BindCraft, BECraft requires a CUDA-compatible Nvidia graphics card to run. In the cuda setting, please specify the CUDA version compatible with your graphics card, for example '11.8'. If unsure, leave blank but it's possible that the installation might select the wrong version, which will lead to errors. In pkg_manager specify whether you are using 'mamba' or 'conda', if left blank it will use 'conda' by default. Note: This install script will install PyRosetta, which requires a license for commercial purposes. The code requires about 2 Mb of storage space, while the AlphaFold2 weights take up about 5.3 Gb.
Next, navigate to the ColabDesign directory and run pip install -e . to install the ptmenergy loss function.
To run binder design, activate the newly created BindCraft environment and navigate to the becraft directory. Then, execute the following command:
python -u ./bindcraft.py --settings './settings_target/PDL1.json' --filters './settings_filters/no_filters.json' --advanced './settings_advanced/default_4stage_multimer.json'
Importantly, in the default configuration provided in default_4stage_multimer.json, the iptmenergy loss has a weight of 1, and the iptm loss has a weight of 0. To use other design configuration jsons, similarly add the iptmenergy field. By default, MPNN optimization is disabled. Please set this to True if necessary.
For further information on all design and filter settings, see the Bindcraft documentation.
In becraft/example, we provide a comparison of 100 designs generated using:
- BindCraft: optimizing ipTM with weight of 1
- BECraft: optimizing interface pTMEnergy with weight of 1
All other parameters are identical for fair comparison.
Binder design outcomes are highly target-dependent. While BECraft generally improves performance, it is not universally superior. Notably, increasing the weight of the energy-based loss (from the default 0.05 in BindCraft to a higher value like 1.0) often yields better results, even when compared to increasing the ipTM loss to the same weight under otherwise identical settings.
As described in our preprint, the pTMEnergy term in BECraft incorporates a weighting scheme derived from the predicted TM-score (pTM) formulation. Specifically, the contribution of each predicted alignment error bin is scaled by a function
However, we sometimes observe that an overly strong emphasis on this local packing, as enforced by the default TM-weighting, can hinder the exploration of other crucial aspects of binder design. In such cases, downweighting or even disabling the TM-weighting might be beneficial.
To address this potential trade-off, BECraft provides the iptmenergy_lambda parameter within the advanced settings JSON file. This parameter serves as a scaling factor for the pTMEnergy loss:
-
Setting
lambdato1(the default value) applies the full calculated pTMEnergy during the design process. This is generally recommended as a starting point to encourage well-packed interfaces. -
Setting
lambdato0completely disables TM-weighting. This results in an energy score that is simply the -LogSumExp over pAE logits, averaged over interface residue pairs.
It is recommended to experiment with different values to determine the optimal balance for a given design problem.