Prediction of hemophilia A severity using a small-input machine-learning framework

Abstract

Hemophilia A is a relatively rare hereditary coagulation disorder caused by a defective F8 gene resulting in a dysfunctional Factor VIII protein (FVIII). This condition impairs the coagulation cascade, and if left untreated, it causes permanent joint damage and poses a risk of fatal intracranial hemorrhage in case of traumatic events. To develop prophylactic therapies with longer half-lives and that do not trigger the development of inhibitory antibodies, it is essential to have a deep understanding of the structure of the FVIII protein. In this study, we explored alternative ways of representing the FVIII protein structure and designed a machine-learning framework to improve the understanding of the relationship between the protein structure and the disease severity. We verified a close agreement between in silico, in vitro and clinical data. Finally, we predicted the severity of all possible mutations in the FVIII structure – including those not yet reported in the medical literature. We identified several hotspots in the FVIII structure where mutations are likely to induce detrimental effects to its activity. The combination of protein structure analysis and machine learning is a powerful approach to predict and understand the effects of mutations on the disease outcome.

Protein residue network analysis reveals fundamental properties of the human coagulation factor VIII

“…In summary, these results indicate that a machine learning framework combined with the FVIII RIN successfully captured the in vitro chromogenic properties of FVIII. Importantly, we could generalize these findings to predict the effect of mutations observed in clinical settings (we report a complete characterization of the relation between the FVIII structure and clinical severity of HA in a separate study 32 ).…”

A Machine Learning Framework Predicts the Clinical Severity of Hemophilia B Caused by Point-Mutations

“…To address this issue, we created an in silico network representation of the FIXa structure—a residue interaction network (RIN)—where each of its residues is a node, and two nodes are connected by an edge if they are in close proximity to each other in the FIXa 3D structure. As we reported previously for FVIII ( Lopes et al, 2021a ; Lopes et al, 2021b ), this novel representation allowed us to calculate several centrality measures of each amino acid, effectively quantifying their importance in the FIXa structure and indicating which amino acids are more or less tolerant to substitutions. To ensure the robustness of this approach, we carefully validated our in silico findings against hundreds of clinical reports associating mutations to the severity of the HB symptoms.…”

F8/F9 variants in the population-based PedNet Registry cohort compared with locus-specific genetic databases of the European Association for Haemophilia and Allied Disorders and the Centers for Disease Control and Prevention Hemophilia A or Hemophilia B Mutation Project

“…For hemophilia B, variants in mild hemophilia B occurred more frequently in the EGF1 domain but less frequently in the pro-peptide and linker than in moderate hemophilia B. The impact of variants on the FVIII or FIX protein structure and consequently on the disease severity is a topic of current research [ [20] , [21] , [22] , [23] ] and evidence is emerging that phenotypic variation may also be related to the region where the variant occurs. [ 20 , 21 ] However, the role of the location of the variant warrants further exploration.…”