arikos/Diet-Driven-Immuno-Metabolic-Score

Machine Learning Proof of Concept for DIMS-AI: A Diet-Driven Immuno-Metabolic Score for Type 2 Diabetes management in Uganda. Combines HbA1c, hs-CRP, and localized dietary data with Explainable AI (SHAP) for low-resource clinical settings. # DIMS-AI: Diet-Driven Immuno-Metabolic Score Machine Learning Proof of Concept for a clinical decision-support tool designed to predict and manage systemic inflammation in Type 2 Diabetes patients within low-resource settings (e.g., Uganda). ## 1. Scientific & Biological Premise **Context:** Current diabetes management heavily relies on HbA1c (glycaemic control), often missing unresolved chronic, low-grade inflammation that drives micro/macrovascular complications. DIMS-AI targets the [...] **Reference Baselines:** The scoring logic adapts principles from the Dietary Inflammatory Index (DII) but is explicitly recalibrated for Sub-Saharan African nutritional transitions. It moves beyond Western-centric baseli[...] ## 2. Mathematical Formulation The composite Immuno-Metabolic Score is derived from a biologically grounded, weighted linear combination of clinical and dietary features, normalized to a 0–100 scale. **Raw Risk Calculation:** Where: - $X_i$ represents the normalized nutritional and clinical feature inputs (e.g., HbA1c, hs-CRP, FFQ ordinal frequencies). - $w_i$ represents the biologically informed inflammatory/metabolic effect weights (positive for risk factors like refined carbs; negative for protective factors like legumes/fermented foods). - $\epsilon \sim \mathcal{N}(0, \sigma^2)$ represents Gaussian stochastic noise, simulating unmeasured biological confounders (e.g., genetics, stress) to prevent model overfitting. **Normalized Composite Score (0–100):** }{%5Cmax(DIMS_{raw})%20-%20%5Cmin(DIMS_{raw})}%5Cright)) ## 3. Pipeline & Software Architecture **Data Sources:** The model is currently validated on a biologically plausible synthetic bio-cohort ($N=1,000$ virtual Ugandan T2D patients). Distribution parameters are strictly informed by regional epidemiological baselines (e.g., mean HbA1c $\approx 9.9\%$ in Ugandan cohorts; elevated baseline hs-CRP distributions in African populations). **Feature Pipeline (11 Inputs):** 1. **Clinical Biom …