Urda-García et al. present a transcriptomics-based analysis of disease co-occurrence using RNA-seq data from 45 human diseases. This study evaluates whether similarities in gene expression profiles can explain known epidemiological comorbidities more effectively than prior network-based approaches.

The authors derived disease-level expression signatures and used these to construct a Disease Similarity Network, in which statistically significant correlations indicate shared molecular patterns. This network reproduced a substantial fraction of known disease co-occurrences. To address disease heterogeneity, patients were further grouped into expression-defined subtypes (“meta-patients”), which were incorporated into a Stratified Similarity Network. This stratified model increased recall of epidemiological associations to 64% and revealed subtype-specific relationships that were not detectable at the disease level, including associations restricted to specific breast cancer subtypes.

Pathway-level analysis using Reactome showed that diseases linked by epidemiological co-occurrence shared significantly more dysregulated pathways than unrelated disease pairs. Immune system pathways were the most consistently shared features, with over 95% of epidemiologically linked disease pairs exhibiting common immune pathway upregulation.

The study demonstrates that patient stratification improves the detection of molecular similarities underlying disease co-occurrence and provides a reproducible framework, supported by a public web resource, for exploring subtype-resolved comorbidity at the transcriptomic level.