Machine-learning driven predictive platform (MDPP) to risk-stratify malaria in pregnancy (MiP)

Malaria in pregnancy (MiP) remains a significant global health challenge which causes substantial maternal morbidity and adverse birth outcomes. The outcomes of MiP are dynamic and depend upon host/parasite derived factors. MiP is often difficult to diagnose and hence is not treated, which affects the mother as well as the offspring. Metabolite biomarkers have been investigated to diagnose MiP, however, there are few that could predict the outcomes of MiP. In this current study we leveraged the diagnostic clinical trial “LAMPREG” and developed a predictive model using machine learning (ML)-based on metabolomics and clinical data that could forecast the outcomes of MiP. In doing so, we stratified MiP patients into high- and low-risk groups and evaluated the differential host/parasite response between malaria infected and uninfected pregnant women. We analyzed whole blood samples from sixty-eight MiP patients with targeted metabolomics/lipidomics using a liquid chromatography/mass spectrometry (LCMS/MS) platform. We identified a panel of 11 metabolites and 4 clinical features that maintained predictive accuracy. For this, a novel multi-method feature importance framework combining Random Forest, LASSO regression, Mutual Information and Gradient Boosting with cross-method concordance for robust feature selection was used. Using these selected features, we trained and compared four ML-algorithms (Support Vector machines, Gradient Boosting, Random Forest and Logistic Regression) for stratifying MiP patients. Comprehensive performance evaluation showed SVM achieved superior overall performance across multiple metrices including highest AUC (0.913), accuracy (0.901), recall (0.907) and F1-score (0.899). Our results demonstrate that interweaving metabolomic and clinical profiles can identify at-risk pregnancies before onset of the disease. This would offer a window for targeted interventions to improve maternal/fetal outcomes in malaria-endemic regions and reveal potential therapeutic targets beyond the current antimalarial arsenal. Our novel proof-of-concept strategy will be a step forward towards precision medicine.