Prediction of inhibitor risk in haemophilia A using machine learning

Introduction and Objective: In Haemophilia A, the development of alloantibodies, known as inhibitors, against infused factor VIII remains a major complication, leading to reduced efficacy and increased morbidity. Inhibitor formation is a multifaceted process influenced by genetic and environmental factors, including the type of mutations in the F8 gene and of the FVIII concentrate, and the patient’s immune system profile. In this study, through the lens of machine learning (ML), we evaluated the possibility to predict inhibitor occurrence in haemophilia A for the frequent missense and nonsense variants, characterised by high and low inhibitor risk, respectively. Material and Methods: We compiled an extensive dataset comprising 1195 missense and 228 nonsense mutations from the genotypic profiles of haemophilia A patients. Using these datasets, we trained various ML models to identify and learn patterns predictive of inhibitor formation. The models were designed to operate without the inclusion of human leukocyte antigen (HLA) type data. Our methodology focused on optimizing the algorithms for high-dimensional, sparse genetic data and ensuring the robustness of predictions through cross-validation techniques. Results: Our ML models, in the absence of any HLA typing information, achieved prediction accuracies of 62% for missense mutations and 72% for nonsense mutations. Importantly, the implementation of these models is open-source, as we aim to facilitate accessibility and collaborative improvements. Conclusions: Our results suggest that ML models can predict inhibitor development with a fair degree of accuracy using only F8 genetic mutation data, which further supports the leading role of F8 for inhibitor risk. As HLA data and additional biomarkers become available, we anticipate that the accuracy of our predictive models will improve further. This innovation holds the promise of becoming a valuable asset in preemptive clinical decision-making, offering the potential to refine treatment strategies before the initiation of therapy, thereby optimizing patient outcomes and advancing the quality of care in haemophilia.