A 20-Gene Whole Blood Transcriptomic Panel for Rapid Diagnosis of Bacterial Infection: Machine Learning Discovery, Bootstrap Stability Validation, and Blind Cross-Population Generalisation

Defensive publication establishing prior art for a 20-gene whole blood transcriptomic diagnostic panel for bacterial infection. Trained on 862 samples across 8 geographically diverse datasets (Southeast Asia, Europe, USA). RF LODO AUC = 0.967 ± 0.046. Blind test AUC = 1.000 on independent Indian and Australian cohorts. Key findings: S100A12 as dominant pan-bacterial marker (bootstrap stability = 1.00); LRRN3 and CACNA2D3 as novel lymphopenia proxy markers (confirmed 9/10 datasets across 5 continents). Permutation test: 0 of 1,000 random panels matched the actual panel (empirical p ≤ 0.001). Released under AGPL v3 — irrevocably open source. No patent may be obtained on any element described herein.