D-SCRIPT

D-SCRIPT predicts physical protein-protein interactions from amino acid sequences using a deep learning model that estimates inter-protein contact maps to enable cross-species generalization and inference from limited experimental data.

Key Features:

• Sequence-Based Interaction Prediction: Predicts protein-protein interactions solely from amino acid sequences without requiring within-species experimental PPI data.
• Structurally-Motivated Design: Estimates an inter-protein contact map as the penultimate stage, reflecting that physical interactions require contacts between specific amino acid subsets.
• Deep Learning Language Modeling: Leverages advances in deep learning language modeling of protein structure to inform interaction prediction.
• Improved Generalizability: Emphasizes structural features to generalize predictions across species.
• Robustness to Limited Training Data: Maintains predictive performance when training data are limited in size or diversity.

Scientific Applications:

• Systems Biology and PPI Network Analysis: Facilitates discovery and functional interpretation of proteins through predicted protein-protein interaction networks.
• Cross-Species Functional Inference: Enables transfer of interaction-based functional characterization across species, exemplified by models trained on human PPIs improving characterization of fly proteins.
• Structural Validation of Predictions: Produces contact maps that align significantly with ground truth from protein complexes with known 3-D structures.

Methodology:

Uses a deep learning model that takes amino acid sequences as input, applies deep learning language modeling of protein structure, and produces an inter-protein contact map as the penultimate output.