PrxteinMPNN Documentation

Run on Colab

Welcome to PrxteinMPNN’s documentation! This project provides a functional interface for ProteinMPNN, leveraging the JAX ecosystem for accelerated computation.

Contents:

• API Reference
  • JacobianSpecification
  • RunSpecification
  • SamplingSpecification
  • ScoringSpecification
  • sample()
  • score()
  • io: Input/Output for Structure and Trajectories
  • mpnn: Fetch model parameters
  • model: MPNN model architecture
  • sampling: Sampling methods for MPNN
  • scoring: Scoring functions for MPNN
  • utils: Utility functions for MPNN
• Tutorials
  • Quick Start
  • Installation Guide
  • Basic Usage
  • Getting Started
  • Advanced Topics
• Examples
  • Advanced Example
  • Basic Examples
  • Advanced Examples
  • Code Examples

Key Features

• Increased Transparency: Clear and functional interface for ProteinMPNN

• JAX Compatibility: Efficient computation with JAX’s functional programming paradigm

• Modular Design: Easy updates and extensions

• Performance Optimization: Utilizes JAX’s JIT compilation and vectorization

• Validated Implementation: >0.95 correlation with ColabDesign across all decoding paths

Validation

PrxteinMPNN has been rigorously validated against the original ColabDesign ProteinMPNN implementation:

Equivalence Test Results

Decoding Path	Correlation	Status
Unconditional	0.984	✅ Validated
Conditional	0.958-0.984	✅ Validated
Autoregressive	0.953-0.970	✅ Validated

All three decoding paths achieve >0.95 Pearson correlation with ColabDesign outputs, ensuring faithful reproduction of the original model’s behavior.

For detailed validation results and testing procedures, see FINAL_VALIDATION_RESULTS.md.

Installation

pip install -e .

For development:

pip install -e ".[dev]"

Use notes