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

Pathogenic responses against self and foreign antigens in systemic autoimmunity and infection, respectively, engage similar immunologic components, thus lacking distinguishing diagnostic biomarkers. Herein, we tested whether whole-blood transcriptome analysis discriminates autoimmune from infectious diseases.

We applied nested cross-validation methodology to tune and validate random forests, k-nearest neighbors, and support vector machines, using a new preprocessing method on 22 publicly available datasets, including 594 patients with a broad spectrum of systemic autoimmune diseases and 615 patients with diverse viral, bacterial, and parasitic infections.

Our preprocessing method tackled RNA sequencing batch effects by sorting the genes within each sample according to individual relative expression values and discriminated between the corresponding pathologies with 98% accuracy versus 63% when using raw values. This model was further tested in external datasets comprising various autoimmune diseases and infections new to its training process, yielding accuracies ranging between 80% and 96%. Enrichment analyses of 457 of the most informative genes identified SAP1, ELF1/4, and FLI1 transcription factors among the significant upstream regulators and revealed several key processes and pathways, such as autophagy, DNA damage response, and NOTCH signaling. A subset of 24 genes, including the inflammation-related genes RPL7, TLK2, and ANK2, distinguished between autoimmune and infectious diseases with 89% accuracy.

Using a novel batch-correction algorithm, this analysis may provide a new mechanistic understanding of the pathogenic autoimmune response, as well as biomarkers for differential diagnoses of the corresponding pathologies in patients presenting with inflammatory disorders.

This work was funded by the European Regional Development Fund NSRF 2014-2020, no. MIS5002802.