Multi-label multi-instance transfer learning for simultaneous reconstruction and cross-talk modeling of multiple...

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Signaling pathways are crucial for cell growth, apoptosis, and development, but current signal transduction networks are incomplete.
• Experimental methods for pathway reconstruction are time-consuming and costly.
• Existing computational methods rarely model multiple signaling pathways simultaneously for novel component discovery and cross-talk analysis.

Purpose of the Study:

• To develop a computational method for simultaneously reconstructing multiple human signaling pathways and modeling their cross-talks.
• To discover novel signaling components and pathway-targeted proteins.
• To provide insights into regulatory and cooperative relationships between signaling pathways.

Main Methods:

• A multi-label multi-instance transfer learning method was proposed.
• Reconstruction of 27 human signaling pathways and their cross-talks.
• Utilized experimentally derived protein-protein interactions (PPIs) and gene ontology enrichment analysis.

Main Results:

• The method achieved satisfactory multi-label learning performance and accurate proteome-wide predictions.
• Predicted signaling components and targeted proteins were validated by recent literature.
• A map of pathway cross-talks was inferred, revealing regulatory and cooperative relationships.

Conclusions:

• The multi-label learning framework effectively models proteins belonging to multiple pathways.
• Novel signaling components and pathway-targeted proteins were predicted.
• The study provides a reconstructed map of human signaling pathways and their cross-talks for future research.