bindcraft

name	bindcraft
description	Computational peptide binder design using BindCraft pipeline (ProteinMPNN inverse folding + Chai/ESMFold structure validation + energy scoring + quality control)
metadata	{"openclaw":{"requires":{"env":["OPENAI_API_KEY"],"bins":["python3"],"anyBins":["proteinmpnn","protein_mpnn"]},"primaryEnv":"OPENAI_API_KEY"}}

BindCraft — Computational Peptide Binder Design

Design peptide binders for target proteins using the BindCraft pipeline. This skill orchestrates inverse folding (ProteinMPNN), structure prediction (Chai/ESMFold), energy scoring, and sequence quality control in an iterative refinement loop.

Capabilities

• Inverse folding (ProteinMPNN): Design amino acid sequences for binder backbones using ProteinMPNN with configurable sampling temperature, batch size, and model weights
• Structure prediction (Chai): Validate designed sequences by folding target+binder complexes and evaluating structural metrics (pAE, ipTM, pLDDT)
• Energy scoring: Score folded complexes using a simplified energy function to rank binder candidates
• Sequence quality control: Filter sequences based on repeat content, charge, hydrophobicity, diversity, and problematic motifs
• Iterative refinement: Run multiple rounds of design→fold→score→filter to converge on high-quality binders

Usage

python skills/bindcraft/scripts/run_bindcraft.py \
  --target-sequence "MKWVTFISLLLLFSSAYSRGV..." \
  --binder-sequence "MKQHKAMIVALIVICITAVVAALVTRKDLCEVHIRTGQTEVAVF" \
  --num-rounds 3 \
  --num-seq 25 \
  --output-dir ./bindcraft_output

Provide a target protein sequence and an initial binder scaffold sequence. The skill will:

1. Run ProteinMPNN inverse folding to generate candidate binder sequences
2. Filter sequences through quality control checks
3. Fold target+binder complexes with Chai structure prediction
4. Score complexes with energy function
5. Repeat for N rounds, refining the best candidates

Parameters

• target_sequence (required): Amino acid sequence of the target protein
• binder_sequence: Initial binder scaffold sequence (default: standard test peptide)
• num_rounds: Number of design-fold-score rounds (default: 3)
• num_seq: Number of sequences per ProteinMPNN batch (default: 25)
• batch_size: ProteinMPNN batch size (default: 250)
• sampling_temp: ProteinMPNN sampling temperature (default: 0.1)
• device: Compute device, e.g. "cuda:0" or "cpu" (default: cuda:0)
• output_dir: Output directory for results and artifacts
• artifact_store: Path to artifact store for DAG integration
• constraints: JSON string of binding constraints (e.g. {"residues_bind": ["R45", "K78"]})

Quality Control Defaults

• max_repeat: Maximum consecutive repeat length (default: 4)
• max_appearance_ratio: Maximum single-residue frequency (default: 0.33)
• max_charge: Maximum net charge (default: 5)
• max_charge_ratio: Maximum charged residue fraction (default: 0.5)
• max_hydrophobic_ratio: Maximum hydrophobic fraction (default: 0.8)
• min_diversity: Minimum unique residue types (default: 8)

Output Artifacts

The skill produces artifacts in the Artifact DAG (Layer 3):

• SEQUENCE artifacts: Designed binder sequences from ProteinMPNN
• STRUCTURE artifacts: Folded complex structures from Chai
• SCORE artifacts: Energy scores and QC metrics for each design
• SCORE_TABLE artifacts: Summary table of all designs ranked by energy