Expanding P-NET, a multi-purpose biologically informed deep learning framework.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics
• Transcriptomics

Background:

• Traditional machine learning models often lack interpretability in complex biological data analysis.
• Leveraging biological pathways can enhance the understanding of genomic and transcriptomic data.
• Deep learning offers powerful tools for biological data analysis but can be a "black box".

Purpose of the Study:

• To present expanded P-NET, a versatile deep learning framework for computational biology.
• To enhance the interpretability and stability of predictions in genomic and transcriptomic tasks.
• To integrate biological pathway information for deeper biological insights.

Main Methods:

• The study utilizes an enhanced P-NET framework, building upon the original P-NET architecture.
• The framework incorporates gene and pathway information for improved predictive modeling.
• Performance was evaluated on genomic and transcriptomic prediction tasks.

Main Results:

• Expanded P-NET achieves competitive performance in genomic and transcriptomic prediction.
• The framework demonstrates enhanced stability and interpretability compared to traditional machine learning models.
• P-NET 2.0 provides valuable insights into complex biological processes through integrated pathway information.

Conclusions:

• Expanded P-NET is a versatile and powerful framework for deep learning in computational biology.
• The integration of biological pathways significantly improves model interpretability and insight generation.
• The publicly available framework facilitates diverse applications in computational biology research.