Complete end-to-end learning from protein feature representation to protein interactome inference

Area of Science:

• Computational Biology
• Proteomics
• Bioinformatics

Background:

• Co-fractionation coupled with mass spectrometry (CF-MS) is vital for mapping protein-protein interactions (PPIs) under physiological conditions.
• Existing CF-MS analysis pipelines struggle with noise, handcrafted features, and limited focus on pairwise interactions, hindering scalability.
• There is a need for advanced computational tools to overcome these limitations in PPI and protein complex analysis.

Purpose of the Study:

• To introduce FREEPII, a unified deep learning framework for accurate and efficient inference of PPIs and protein complexes.
• To integrate CF-MS data with sequence-derived features for enhanced protein-level representations.
• To develop a scalable and generalizable method for analyzing protein interaction networks.

Main Methods:

• Developed FREEPII, a deep learning framework utilizing a convolutional neural network architecture.
• Integrated raw CF-MS data with sequence-derived features as auxiliary input.
• Employed supervised protein embeddings to encode network-level context from complex annotations.

Main Results:

• FREEPII outperforms state-of-the-art CF-MS analysis tools in capturing biologically coherent protein features.
• The framework demonstrates enhanced robustness against experimental noise by integrating multimodal data.
• Cross-dataset evaluations confirm improved generalization and sensitivity for data-driven PPI inference across species.

Conclusions:

• FREEPII offers a unified computational framework for learning discriminative protein representations from CF-MS and sequence data.
• The deep learning architecture enables accurate, scalable inference of PPIs and protein complexes across species.
• FREEPII provides a flexible foundation for discovering novel protein interactions and exploring protein networks.