martinpacesa/BindCraft

What it solves

BindCraft is a pipeline for de novo binder design, allowing researchers to create proteins that bind to a specific target protein. It automates the process of generating, optimizing, and filtering potential binders to identify candidates for experimental validation.

How it works

The pipeline combines several computational tools to design binders:

• AlphaFold2 (AF2) Backpropagation: Used to optimize the binder sequence and structure based on design weights (such as pLDDT, PAE, and contact numbers).
• ProteinMPNN: Used to generate and optimize sequences that fit the designed backbone structure.
• PyRosetta: Used for final structure relaxation and scoring (e.g., energy scores, shape complementarity, and dG).

Users specify a target protein (PDB file) and optional hotspots (specific residues to target). The system then runs multiple trajectories using various design algorithms (e.g., 4stage, greedy, MCMC) to evolve the binder. The resulting designs are then passed through a series of filters based on AF2 confidence scores and Rosetta energy metrics to select the best candidates.

Who it’s for

It is intended for computational biologists and protein engineers who want to design high-affinity binders for specific protein targets.

Highlights

• Automated Pipeline: Handles everything from initial design to final filtering.
• Flexible Design Algorithms: Offers multiple strategies for sequence optimization (2stage, 3stage, 4stage, greedy, MCMC).
• Cuda-Compatible: Optimized for NVIDIA GPUs to handle the large memory requirements of AF2.
• Extensive Filtering: Provides a wide array of metrics (pLDDT, i_pTM, Rosetta energy, SASA) to ensure high-quality designs.
• Target Site Selection: Can either use user-defined hotspots or allow AF2 to automatically detect binding sites.