Human Pan-Disease Whole Blood Transcriptome Atlas: Machine Learning Reveals Cross-Disease Systemic Features

This article constructs the Human Pan-Disease Whole Blood Transcriptome Atlas (WBT), using machine learning to analyze whole blood RNA-seq data from 4,444 samples across 98 diseases, revealing cross-disease systemic gene expression features. This study marks a shift in disease research from an isolated perspective to a systemic one, providing new directions for precision medicine, biomarker development, and therapeutic target exploration.

Whole blood has advantages such as systemic representation, accessibility, dynamism, and clinical relevance; its transcriptome can reflect the physiological and pathological state of the entire body. The transcriptome provides a snapshot of gene expression activity, and compared to the genome, it more directly reflects functional status, dynamic responses, and regulatory insights, making it an ideal subject for studying the systemic mechanisms of diseases.

The WBT integrates RNA-seq data from 4,444 samples across 98 diseases, using batch correction methods like ComBat to eliminate technical variations, and addresses heterogeneity issues through a unified data processing pipeline. The core analysis framework is based on machine learning, including classification models, feature selection, cluster analysis, and network analysis, to identify disease-related transcriptome features.

The WBT reveals cross-disease shared features (such as inflammatory pathway activation, immune regulation disorders, metabolic reprogramming, and cellular stress responses) and disease-specific features (unique gene expression patterns, regulatory network remodeling, and severity markers). Additionally, multi-omics integration (genomics, proteomics, metabolomics) provides more comprehensive insights into molecular mechanisms.

The transcriptome features identified by WBT can be used to develop diagnostic biomarkers (disease classification, early detection, differential diagnosis), guide therapeutic target discovery (shared targets, precision medicine, drug repurposing), and support disease monitoring and prognosis (treatment response, recurrence prediction, complication risk assessment).

The current WBT has limitations such as tissue specificity, insufficient temporal dimension, unvalidated causal inference, and population representativeness bias. Future directions include developing longitudinal cohorts, single-cell/spatial transcriptomics, functional validation, and clinical translation to refine the study of disease molecular mechanisms.

The WBT demonstrates the transformative potential of AI and big data in medicine: new models of data-driven discovery, systemic insights from cross-study integration, AI-assisted mechanism understanding (requiring human experimental validation), and the importance of balancing ethics and privacy.