DeepHisCoM: deep learning pathway analysis using hierarchical structural component models.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Traditional pathway analysis methods often overlook nonlinear interactions between biological factors (genes, proteins) and pathways.
• Identifying disease-associated pathways is crucial for understanding disease mechanisms and developing targeted therapies.

Purpose of the Study:

• To introduce Deep-learning pathway analysis using Hierarchical structured CoMponent models (DeepHisCoM), a novel deep learning framework.
• To address the limitations of existing methods by incorporating complex nonlinear relationships between biological factors and pathways.
• To evaluate DeepHisCoM's performance against conventional methods using simulations and real-world omics data.

Main Methods:

• Developed DeepHisCoM, a multilayered deep learning model that accounts for hierarchical biological structures.
• Utilized deep learning to model nonlinear contributions of biological factors to pathways.
• Validated the model through simulation studies and application to hepatocellular carcinoma (HCC) and coronavirus disease-2019 (COVID-19) datasets.

Main Results:

• DeepHisCoM demonstrated higher statistical power for detecting nonlinear pathway effects compared to conventional methods.
• The method showed comparable power for linear pathway effects.
• Applied to HCC omics data, DeepHisCoM identified key pathways including lysine degradation and amino acid biosynthesis.
• Analysis of COVID-19 SNP data revealed associations with pathways like MAPK signaling and cardiomyopathy.

Conclusions:

• DeepHisCoM offers a powerful new approach for pathway analysis, particularly excelling in capturing nonlinear biological relationships.
• The method successfully identified disease-associated pathways in complex diseases like HCC and COVID-19.
• DeepHisCoM has the potential to advance our understanding of disease biology and biomarker discovery.