ML-powered outbreak prediction for all 37 Nigerian states. Combines NCDC epidemiological surveillance data, ERA5 climate reanalysis, and antiviral drug discovery — all updated weekly.
Binary classification: will a state report >5 confirmed Lassa cases in the next 4 weeks? Trained with strict temporal cross-validation — no data leakage between train/validate/test splits.
Top features: 8-week rolling case mean (49.6%), 1-week lag cases (26.1%), 4-week rolling mean (13.6%), dry season timing (3.5% combined). Weather variables (rainfall, humidity, temperature) contribute <3% — recent case history dominates, consistent with Lassa's focal rodent-reservoir transmission ecology.
Limitations: Confirmed-case counts under-report true incidence — Lassa is difficult to diagnose clinically and PCR access is uneven across states — so both the prediction target and the dominant case-history features inherit surveillance bias. Ideal predictors such as rodent-reservoir ecology are not yet available at scale, so rainfall serves only as a partial environmental surrogate. Case data 2012–2015 are seed estimates (not direct NCDC surveillance counts). Weather is from ERA5 reanalysis at state capital coordinates — intra-state variation not captured. Model should be treated as a screening tool, not a replacement for NCDC field surveillance.
Enter LASV protein targets into GaiaLab's AI analysis pipeline. The viral pathogen mode routes GPC, NP, L, and Z through a Lassa-specific drug repurposing engine with known antiviral candidates pre-loaded.