Extracting a COVID-19 signature from a multi-omic dataset

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• COVID-19 complexity necessitates advanced diagnostic approaches beyond symptom tracking.
• Multi-omic data integration (clinical, proteomic, metabolomic) offers deeper insights into disease mechanisms and biomarker discovery.

Purpose of the Study:

• To develop a machine learning framework for classifying COVID-19 status using multi-omic data.
• To identify condensed, interpretable biomarker signatures for COVID-19.

Main Methods:

• Collected extensive clinical, proteomic, and metabolomic datasets from COVID-19 positive and negative patients.
• Employed a multi-view machine learning framework with ensemble methods to integrate thousands of features.
• Utilized a novel feature relevance methodology for signature identification.

Main Results:

• Achieved a balanced accuracy of 89% ± 5% in COVID-19 classification.
• Identified 12- and 50-feature signatures that improved classification accuracy by at least 3% over the full dataset.
• Demonstrated the accuracy and interpretability of the derived signatures.

Conclusions:

• Multi-omic data integration and machine learning effectively extract robust COVID-19 signatures.
• Condensed biomarker sets offer a practical pathway for improved diagnosis and precision medicine.
• This work represents a significant advancement in COVID-19 biomarker discovery.