Coagulation Factor X (FX), often termed Stuart-Prower Factor, is a plasma glycoprotein composed of the γ-carboxyglutamic acid (Gla) domain, two epidermal growth factor domains (EGF-1, EGF-2) and the serine protease (SP) domain. FX plays a pivotal role in the coagulation cascade, activating thrombin to promote platelet plug formation and prevent excess blood loss. Genetic variants in FX disrupt coagulation and lead to FX or Stuart-Prower Factor deficiency. To better understand the relationship between FX deficiency and disease severity, an interactive FX variant database has been set up at https://www.factorx-db.org, based on earlier websites for the Factor XI and IX coagulation proteins. To date (April 2021), we report 427 case reports on FX deficiency corresponding to 180 distinct F10 genetic variants. Of these, 149 are point variants (of which 128 are missense), 22 are deletions, three are insertions and six are polymorphisms. FX variants are phenotypically classified as being Type I or Type II. Type I variants involve the simultaneous reduction of FX coagulant activity (FX:C) and FX antigen levels (FX:Ag), whereas Type II variants involve a reduction in FX:C with normal FX:Ag plasma levels. Both types of variants were distributed throughout the FXa protein structure. Analyses based on residue surface accessibilities showed the most damaging variants to occur at residues with low accessibilities. The interactive FX web database provides a novel easy-to-use resource for clinicians and scientists to improve the understanding of FX deficiency. Guidelines are provided for clinicians who wish to use the database for diagnostic purposes.

Coagulation Factor XI (FXI) is a plasma glycoprotein composed of four apple (Ap) domains and a serine protease (SP) domain. FXI circulates as a dimer and activates Factor IX (FIX), promoting thrombin production and preventing excess blood loss. Genetic variants that degrade FXI structure and function often lead to bleeding diatheses, commonly termed FXI deficiency. The first interactive FXI variant database underwent initial development in 2003 at https://www.factorxi.org. Here, based on a much improved FXI crystal structure, the upgraded FXI database contains information regarding 272 FXI variants (including 154 missense variants) found in 657 patients, this being a significant increase from the 183 variants identified in the 2009 update. Type I variants involve the simultaneous reduction of FXI coagulant activity (FXI:C) and FXI antigen levels (FXI:Ag), whereas Type II variants result in decreased FXI:C yet normal FXI:Ag. The database updates now highlight the predominance of Type I variants in FXI. Analysis in terms of a consensus Ap domain revealed the near-uniform distribution of 81 missense variants across the Ap domains. A further 66 missense variants were identified in the SP domain, showing that all regions of the FXI protein were important for function. The variants clarified the critical importance of changes in surface solvent accessibility, as well as those of cysteine residues and the dimer interface. Guidelines are provided below for clinicians who wish to use the database for diagnostic purposes. In conclusion, the updated database provides an easy-to-use web resource on FXI deficiency for clinicians.

Computational approaches for predicting the pathogenicity of genetic variants have advanced in recent years. These methods enable researchers to determine the possible clinical impact of rare and novel variants. Historically these prediction methods used hand-crafted features based on structural, evolutionary, or physiochemical properties of the variant. In this study we propose a novel framework that leverages the power of pre-trained protein language models to predict variant pathogenicity. We show that our approach VariPred (Variant impact Predictor) outperforms current state-of-the-art methods by using an end-to-end model that only requires the protein sequence as input. By exploiting one of the best-performing protein language models (ESM-1b), we established a robust classifier, VariPred, requiring no pre-calculation of structural features or multiple sequence alignments. We compared the performance of VariPred with other representative models including 3Cnet, Polyphen-2, FATHMM and ‘ESM variant’. VariPred outperformed all these methods on the ClinVar dataset achieving an MCC of 0.727 vs. an MCC of 0.687 for the next closest predictor.