Understanding the molecular foundations of disease progression remains a central challenge in medicine. While longitudinal health records reveal large-scale patterns of disease trajectories, their molecular interpretation has been limited. Here, we present a unified framework that integrates disease-gene associations from eight major resources, including OpenTargets, DisGeNET, and OMIM, into the human protein-protein interaction network and links them to empirical disease trajectories derived from 44 million hospitalizations in Austria.

The resulting dataset provides molecular profiles for 1,023 ICD-10 disease codes encompassing 18,897 proteins. Diseases within the same ICD chapters showed high molecular similarity, which converged with their phenotypic and epidemiological closeness. Network centrality of disease-associated proteins in the interactome was predictive for disease progression, distinguishing favorable from adverse health trajectories. Integration with drug-target data further identified trajectory-specific therapeutic opportunities, validated through literature evidence and AI agent-based predictions.

By coupling molecular interaction networks with population-scale health trajectories, this work provides a systematic molecular framework for understanding disease co-occurrence and progression, opening new avenues for precision prevention, early intervention, and drug repurposing. All results are accessible via an interactive web application (https://vis.csh.ac.at/disease-trajectories/), enabling open exploration of the molecular atlas of disease trajectories.