Transfer transcriptomic signatures for infectious diseases.

Area of Science:

• Computational Biology and Bioinformatics
• Immunology
• Infectious Diseases

Background:

• Transcriptomic signatures are early indicators of infection but are limited by clinical study constraints, impacting their generalizability.
• Existing transcriptomic signatures often perform poorly on new datasets due to sample limitations and disease specificity.

Purpose of the Study:

• To develop a machine learning approach for identifying common transcriptomic elements across diverse diseases.
• To create generalizable gene sets, termed 'transfer signatures,' predictive of disease states across different datasets and species.

Main Methods:

• Utilized a machine learning framework to identify common gene expression patterns within and across distinct biological datasets.
• Empirically validated the predictive power of identified gene sets (transfer signatures) on independent datasets from different diseases.
• Applied transfer signatures to analyze tuberculosis progression and COVID-19/influenza A H1N1 infection severity.

Main Results:

• Identified 'transfer signatures' — sets of genes consistently predictive across diverse datasets and species (e.g., rhesus to human).
• Demonstrated the utility of transfer signatures in predicting tuberculosis progression and severity of viral infections like COVID-19 and influenza.
• Showcased the ability of transfer signatures to function effectively in the absence of disease-specific biomarkers.

Conclusions:

• Transfer signatures offer a robust, generalizable method for identifying disease states from transcriptomic data.
• A small set of archetypal human immunophenotypes, captured by transfer signatures, can explain broad responses to diverse diseases.
• This approach has significant implications for biomarker discovery and clinical applications, especially in resource-limited settings.