Computational binder design has become one of the most productive applications of protein structure prediction. BindCraft, developed by Martin Pacesa, combines AlphaFold2-based backbone generation with ProteinMPNN sequence design and multi-stage relaxation to produce candidate binders against arbitrary protein targets. The pipeline works. The problem is access: BindCraft depends on PyRosetta for structural relaxation, shape complementarity scoring, and solvent-accessible surface area calculations. PyRosetta requires a license that excludes many commercial and industrial users, and even academic groups sometimes encounter delays or restrictions that interrupt ongoing work.
FreeBindCraft addresses this constraint directly. It is a fork of BindCraft (v1.52) that makes PyRosetta optional by replacing each PyRosetta-dependent calculation with an open-source alternative. The design pipeline remains functionally equivalent while removing the licensing barrier entirely.
Replacing PyRosetta: component by component
The PyRosetta dependency in BindCraft serves four distinct functions: structural relaxation, shape complementarity scoring, SASA calculation, and side-chain repacking. FreeBindCraft replaces each one.
Structural relaxation with OpenMM
The most computationally intensive PyRosetta operation in BindCraft is FastRelax, which energy-minimizes designed structures through repeated cycles of repacking and gradient descent. FreeBindCraft replaces this with an OpenMM-based protocol that includes structure preparation via PDBFixer, ramped backbone restraints, OBC2 implicit solvation, and an additional short-range repulsive term to reduce clashes. Early relaxation stages include short molecular dynamics "shakes" that help escape local energy minima.
The OpenMM protocol runs on GPU hardware and is typically 2 to 4 times faster than CPU-based FastRelax. Because relaxation is the rate-limiting step in the BindCraft pipeline, this acceleration compounds across the dozens or hundreds of candidate structures generated in a typical design campaign.
Shape complementarity via sc-rs
Shape complementarity (SC) quantifies geometric fit between a designed binder and its target surface. PyRosetta computes this using the Lawrence and Colman algorithm. FreeBindCraft uses sc-rs, an open-source Rust implementation of the same algorithm. Benchmarking shows that sc-rs produces values nearly identical to PyRosetta's SC calculations across diverse interfaces, while running as a standalone binary with no Python runtime overhead.
SASA and interface metrics
Solvent-accessible surface area calculations, which BindCraft uses to evaluate buried surface area at the interface, are handled by FreeSASA with a Biopython Shrake-Rupley fallback. Interface residue identification, RMSD calculations, and structural alignment use Biopython routines. These values align closely with their PyRosetta-derived equivalents, and where Rosetta-specific metrics lack open-source counterparts, the pipeline uses placeholder values that are compatible with the default filter architecture.
Side-chain repacking with FASPR
FASPR (Fast, Accurate, and Deterministic Protein Side-chain Packing) handles side-chain optimization within the relaxation protocol. It is integrated into the OpenMM workflow so that repacking and energy minimization alternate in a manner analogous to the original FastRelax protocol.
ipSAE scoring
FreeBindCraft also introduces support for ipSAE (interface predicted Structural Alignment Error), a metric based on Dunbrack et al. (2025) that evaluates interface quality using AlphaFold's PAE matrix. Unlike i_pTM, which uses a global distance normalization, ipSAE applies per-residue normalization based on the number of interface contacts. This provides a complementary ranking signal: users can choose to rank final designs by either i_pTM or ipSAE depending on which metric better captures their design objectives.
Design modes
FreeBindCraft supports two design modes. Miniprotein binders (typically 65 to 150 residues) use the standard multi-stage BindCraft pipeline with default or relaxed filters. Peptide binders (8 to 30 residues) use peptide-specific filters and advanced profiles tuned for shorter sequences where backbone flexibility and interface geometry differ substantially from structured miniproteins.
CogniChem integration
On CogniChem, FreeBindCraft runs as a managed cloud GPU job. Users upload a target PDB, select design mode (miniprotein or peptide), and choose a GPU tier appropriate for their throughput needs. The platform handles environment setup, GPU allocation, and artifact management. Completed designs are delivered as ranked PDB files to the Job Queue, where users can inspect scores, download structures, or pass results directly into downstream analysis.
FreeBindCraft is also available as a node in CogniChem's workflow builder. This means a binder design campaign can be chained with upstream target preparation (e.g., structure prediction or pocket identification) and downstream evaluation (e.g., molecular dynamics or docking) without manual file transfers between steps. The workflow system handles data routing and scheduling across jobs automatically.
What this enables
The practical consequence of removing PyRosetta is that binder design becomes accessible to any research group or company without navigating a separate licensing process. The GPU-accelerated relaxation makes individual design campaigns faster, while the cloud deployment on CogniChem removes the local hardware and environment management requirements that make BindCraft difficult to set up from scratch.
Together, these changes convert binder design from a tool that requires specific institutional access and local GPU infrastructure into a service that any computational biologist can use immediately. The underlying design algorithm remains BindCraft's validated pipeline; what changes is who can run it and how quickly they can iterate.
To try FreeBindCraft, visit the CogniChem toolkit or create an account to submit your first design job.