APNet, an explainable sparse deep learning model to discover differentially active drivers of severe COVID-19

Area of Science:

• Computational biology
• Translational medicine
• Proteomics

Background:

• Omics data analysis offers insights into complex diseases like COVID-19.
• Current computational methods often miss nonlinear proteomic dynamics and lack interpretability.

Purpose of the Study:

• To introduce APNet, a novel pipeline for explainable COVID-19 severity prediction.
• To improve biological interpretability beyond feature ranking in omics analyses.

Main Methods:

• APNet combines differential activity analysis (SJARACNe networks) with a deep learning model (PASNet).
• It utilizes driver-pathway networks for result interpretation and hypothesis generation.

Main Results:

• APNet accurately classifies COVID-19 patient severity across three proteomic datasets.
• It identifies predictive drivers and pathways, some validated by single-cell omics data.

Conclusions:

• APNet offers a powerful, interpretable approach for COVID-19 severity prediction using omics data.
• The pipeline highlights novel biomarker circuitries for further investigation in COVID-19 research.