Topsy-Turvy

Topsy-Turvy predicts protein-protein interactions by integrating sequence-based multi-scale deep-learning with global network patterns to enable genome-scale and cross-species PPI prediction.

Key Features:

• Integration of Sequence-Based and Global Insights: Uses transfer-learning to combine bottom-up sequence-derived molecular features and top-down global interaction patterns during training.
• Cross-Species Genome-Scale Prediction: Enables genome-scale PPI predictions for non-model organisms that lack experimental PPI data.
• Hybrid Model (TT-Hybrid): Integrates Topsy-Turvy sequence-based predictions with a network-based link prediction approach to model species-specific network rewiring and predict interactions for well- and sparsely-characterized proteins.
• Scalability Compared to AlphaFold-Multimer: Is designed to scale efficiently across whole genomes, making it feasible for large-scale screenings relative to AlphaFold-Multimer.
• Accuracy and Generalizability: Delivers highly accurate predictions and generalizes across species, outperforming other state-of-the-art PPI prediction methods.

Scientific Applications:

• Genome-wide PPI mapping: Provides interpretable PPI predictions at the genome level for model and non-model organisms.
• Evolutionary biology: Enables comparative studies of protein interaction evolution across species.
• Functional genomics: Supports functional annotation and network analysis where experimental PPI data are sparse.
• Network rewiring analysis: Uses TT-Hybrid to investigate species-specific network dynamics and rewiring.

Methodology:

Employs a sequence-based, multi-scale deep-learning model trained with transfer-learning to integrate bottom-up sequence features and top-down global interaction patterns; TT-Hybrid combines sequence-based predictions with network-based link prediction.