Answers are grounded in this paper and cite their sources.
Background and objectives Hemoglobinopathies, including beta thalassemia, are blood disorders caused by mutations in the HBB gene. Accurate and early screening of heterozygous carriers is necessary to prevent mutations from being transmitted to offspring. Diagnosis of carrier status is based on hemoglobin A2 (HbA2) and fetal hemoglobin (HbF) levels, requiring expensive equipment for quantification. This work aims to develop machine learning models for prediction of HbA2 and HbF from routine blood test parameters in persons with hemoglobinopathies and normal individuals. Methods Mixture of Experts (MoE) approach was applied to develop the models using an Indian dataset of 5314 individuals. Post-hoc interpretations were obtained with SHapley Additive exPlanations (SHAP). An affinity-based classifier was developed to predict the diagnostic labels from the predicted HbA2 and HbF. Finally, a web application, HemoPred, was developed as frontend of the models. Results MoE models achieved mean absolute errors of 0.1654 and 0.1394, and R 2 -scores of 0.8442 and 0.6055 for HbA2 and HbF prediction, respectively, with 5-fold cross-validation. The models achieved similar performance in held-out internal validation. SHAP revealed MCV, MCH, and RDW-CV as top contributors. Affinity-based classifier achieved 95% top-2-accuracy (5-fold cross-validation) in multiclass prediction. In binary (disease/normal) classification, it achieved 77% balanced accuracy with 93% sensitivity. HemoPred is available at https://aihemo.mp.aau.dk/hemopred . Conclusions ML models developed for HbA2 and HbF prediction, and diagnostic classification from the predicted HbA2 and HbF can aid doctors. This work is limited by unavailable multi-center data for external validation and unmeasured confounders, including nutritional profiles and α-thalassemia co-inheritance status.
No takes yet. Share an insight, caveat, or question.
Bhattacharjee et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: