Machine learning-based integration develops an immune-derived signature for diagnosing high-altitude pulmonary...

High-altitude pulmonary hypertension (HAPH) is a common disease in high-altitude regions where implementation of gold-standard diagnostic methods remains logistically challenging.

In the retrospective analysis, we employed an integrative multi-omics approach combining single-cell RNA sequencing (scRNA-seq, n = 10), bulk RNA sequencing (RNA-seq, n = 126), and proteomic profiling (n = 42) to characterize immune microenvironment remodeling in HAPH. Subsequently, we established a machine learning-based diagnostic model. The HAPH-associated signatures were finally validated by Quantitative PCR.

Through scRNA-seq analysis utilizing Ro/e and contribution scoring analysis, we first demonstrated the pivotal role of myeloid lineages in HAPH pathogenesis. Pseudotime trajectory analysis of the myeloid subsets further revealed 2,615 differentially expressed genes (DEGs) associated with HAPH progression. We also identified 144 and 77 DEGs from bulk RNA-seq and proteomic data between HAPH and control groups, respectively. Finally, 22 candidate biomarkers were screened by muti-omics analysis. These genes were further refined through ensemble machine learning algorithms. Evaluation of 113 algorithm combinations revealed that a six-gene random forest (RF) model (HEMGN, HBG2, MYL9, ANK1, UBE2O, RBPMS2) achieved optimal diagnostic accuracy, with an area under the curve (AUC) of 0.995 in the training cohort (n = 55) and 0.773 in external validation cohorts (n = 71). Quantitative PCR validated significant overexpression of these biomarkers in HAPH compared to controls (p < 0.05).

Our findings propose the minimally invasive blood-derived immune signature for HAPH diagnosis, providing a practical framework for early detection in resource-constrained high-altitude populations.