Machine learning-based identification of a transcriptomic blood signature discriminating between systemic...

Area of Science:

• Immunology and Computational Biology
• Transcriptomics and Bioinformatics

Background:

• Systemic autoimmunity and infections share immunologic components, hindering diagnostic biomarker development.
• Distinguishing between autoimmune and infectious diseases is clinically challenging due to overlapping pathologies.

Purpose of the Study:

• To evaluate whole-blood transcriptome analysis for discriminating autoimmune diseases from infections.
• To develop and validate a novel preprocessing method for RNA sequencing data to improve diagnostic accuracy.

Main Methods:

• Utilized nested cross-validation with machine learning algorithms (random forests, k-NN, SVM) on 22 public datasets (594 autoimmune, 615 infectious diseases).
• Implemented a new preprocessing method to address RNA sequencing batch effects by sorting gene expression values.
• Performed enrichment analyses on informative genes to identify upstream regulators and biological pathways.

Main Results:

• The novel preprocessing method achieved 98% accuracy in discriminating pathologies, compared to 63% with raw data.
• External validation demonstrated accuracies between 80% and 96% for distinguishing new autoimmune and infectious disease datasets.
• A 24-gene signature, including RPL7, TLK2, and ANK2, differentiated autoimmune from infectious diseases with 89% accuracy.

Conclusions:

• A novel batch-correction algorithm applied to whole-blood transcriptomes can effectively differentiate autoimmune from infectious diseases.
• This approach offers potential mechanistic insights into autoimmune pathogenesis and provides novel biomarkers for differential diagnosis.
• The findings support transcriptome analysis as a valuable tool for diagnosing patients with inflammatory disorders.