A feature extraction free approach for protein interactome inference from co-elution data

Area of Science:

• Proteomics
• Systems Biology
• Computational Biology

Background:

• Protein complexes are essential for cellular functions.
• Co-fractionation coupled with mass spectrometry (CF-MS) is a high-throughput method for studying protein complexes.
• Existing computational methods for CF-MS data analysis face challenges with false positives, feature bias, and imbalanced data.

Purpose of the Study:

• To develop a novel computational framework for accurate protein-protein interaction (PPI) and protein complex prediction from CF-MS data.
• To overcome limitations of existing methods, including handcrafted feature bias and imbalanced data issues.
• To improve the sensitivity and reliability of interactome inference.

Main Methods:

• Development of SPIFFED (Software for Prediction of Interactome with Feature-extraction Free Elution Data), an end-to-end deep learning architecture.
• Integration of raw CF-MS elution data using convolutional neural networks.
• Implementation of balanced training strategies and ensemble methods for improved prediction.
• Utilizing ClusterONE for high-confidence protein complex inference.

Main Results:

• SPIFFED outperforms state-of-the-art methods in PPI prediction, especially under imbalanced training conditions.
• Training SPIFFED with balanced data significantly enhances sensitivity for true PPIs.
• Ensemble SPIFFED models provide robust integration of multiple CF-MS datasets.
• The software facilitates the inference of high-confidence protein complexes tailored to experimental designs.

Conclusions:

• SPIFFED offers a powerful, bias-free, and accurate approach for analyzing CF-MS data.
• The feature-extraction-free deep learning architecture effectively predicts protein-protein interactions and complexes.
• SPIFFED represents a significant advancement in computational proteomics for understanding cellular machinery.