PLA-MoRe

PLA-MoRe predicts protein-ligand binding affinities by integrating structural molecular representations (extending beyond SMILES) with multisource bioactive features to improve binding-affinity estimation for drug-discovery applications.

Key Features:

• Dual Representation Approach: Uses two compound feature extractors to combine structural and bioactive information for molecules.
• Structure Feature Extractor (Graph Isomorphism Network): Employs a graph isomorphism network (GIN) to capture molecular graph representations.
• Autoencoder-based Bioactive Feature Extractor: Integrates multisource bioactive information including chemical properties, target interactions, biological networks, cellular contexts, and clinical data via an autoencoder.
• Sequence Feature Extraction: Learns embeddings for protein sequences to represent protein features for interaction modeling.
• Integrated Prediction Framework: Concatenates outputs from the three extractors and processes them through a fully connected network to predict binding affinities.
• Attention Visualization: Produces attention-based visualizations to help locate binding sites and interpret interaction regions.

Scientific Applications:

• Drug Discovery Enhancement: Enables prioritization of potential drug candidates by providing predicted binding affinities to guide experimental validation.
• Mechanistic Insights: Uses attention visualizations to identify putative binding sites and inform hypotheses about interaction mechanisms.

Methodology:

PLA-MoRe was compared against three state-of-the-art methods and subjected to an ablation study to assess the contribution of each model component.