Skip to content

Home

A Hierarchical Multimodal Deep Learning Framework for Accurate and Interpretable Protein-Protein Interaction (PPI) Prediction.

Install ProMaya NextFlow Quick Start

About

ProMaya is a state-of-the-art computational framework designed to accurately predict Protein-Protein Interactions (PPIs) by modeling both sequence and structural features. It fuses multi-scale biological information—ranging from precise atomic electron density (via our novel LSMD metric) to global protein structure and deep language model embeddings—into a unified deep learning architecture.

ProMaya Pipeline Architecture
ProMaya data processing and Nextflow.

Capabilities

ProMaya directly addresses the multi-scale challenges of structural bioinformatics through a robust Hierarchical Multimodal Deep Learning approach:

  • Robust Processing Pipeline: ProMaya is orchestrated by a highly scalable Nextflow pipeline, enabling seamless transition from local single-pair inference to massive HPC cluster screening campaigns.
  • Unprecedented Interpretability: Through a novel application of Graph Grad-CAM and SHAP analysis, ProMaya provides attribution at multiple scales (Residue, Atom, and Surface), eliminating the "black-box" nature of traditional deep learning tools.
  • Multi-Modal Data Fusion:
    1. Atomic Graph: Captures explicit electron density proxies at binding interfaces.
    2. Residue Graph: Fuses biochemical properties, ProtTrans-650M language model embeddings, and intrinsic disorder metrics.
    3. Surface Point Cloud: Models 3D geometric shape and electrostatic complementarity.
    4. Sequence Context: Leverages deep evolutionary information from PSSMs.

Download & Source Code

Source code and the dockerized standalone package of ProMaya are available at zenodo-10-link. Download the model and Docker image from there, then run the Nextflow standalone pipeline.