Gene expression integration and similarity score based modeling improve risk stratification in idiopathic venous thrombophilia

TY - JOUR

T1 - Gene expression integration and similarity score based modeling improve risk stratification in idiopathic venous thrombophilia

AU - Ezquerra-Condeminas, Pol

AU - Martinez-Perez, Angel

AU - Howald, Cedric

AU - Brown, Andrew A

AU - Souto, Juan Carlos

AU - Viñuela, Ana

AU - Buil, Alfonso

AU - Perera-Lluna, Alexandre

AU - Soria, Jose Manuel

N1 - Copyright © 2026. Published by Elsevier Inc.

PY - 2026/5/20

Y1 - 2026/5/20

N2 - BACKGROUND: Idiopathic venous thromboembolism (VTE) occurs in the absence of provoking factors, limiting the efficacy of current risk stratification. In parallel, the lack of integration between transcriptomic data and established risk factors prevents the identification of individuals with a high baseline predisposition.OBJECTIVES: To improve risk stratification of idiopathic VTE beyond traditional clinical models by developing a similarity-based risk score that integrates transcriptomic profiles with conventional risk factors.PATIENTS/METHODS: We analyzed 790 individuals from the GAIT2 familial study, including 70 participants with prior idiopathic VTE. Whole-blood RNA sequencing, known genetic variants, and clinical variables were integrated using supervised machine learning models (Elastic Net and XGBoost). Predictive gene expression features were evaluated through enrichment analyses. A unified similarity score combining both models was developed to identify control individuals who shared transcriptomic and clinical profiles with VTE cases.RESULTS: In both models, von Willebrand factor abundance was the strongest predictor of VTE, followed by clinical factors (BMI, ABO alleles, age) and expression of 494 genes, including STS, FAM13A, GPRIN1, FLVCR2, FAM177B, and several long non-coding RNAs not previously linked to thrombosis. Known thrombosis-associated genes such as UQCRC2 and PRKRA were also identified. Significant enrichment was observed for cardiomyopathic KEGG pathways and renal HPA terms. Similarity-based risk score construction improved classification, with 74% of VTE cases and 23% of controls assigned to the risk zone.CONCLUSIONS: Multivariate integration via machine learning enhances VTE risk stratification, identifying novel transcriptomic signatures and lncRNA biomarkers that offer new strategies for VTE personalized prevention.

AB - BACKGROUND: Idiopathic venous thromboembolism (VTE) occurs in the absence of provoking factors, limiting the efficacy of current risk stratification. In parallel, the lack of integration between transcriptomic data and established risk factors prevents the identification of individuals with a high baseline predisposition.OBJECTIVES: To improve risk stratification of idiopathic VTE beyond traditional clinical models by developing a similarity-based risk score that integrates transcriptomic profiles with conventional risk factors.PATIENTS/METHODS: We analyzed 790 individuals from the GAIT2 familial study, including 70 participants with prior idiopathic VTE. Whole-blood RNA sequencing, known genetic variants, and clinical variables were integrated using supervised machine learning models (Elastic Net and XGBoost). Predictive gene expression features were evaluated through enrichment analyses. A unified similarity score combining both models was developed to identify control individuals who shared transcriptomic and clinical profiles with VTE cases.RESULTS: In both models, von Willebrand factor abundance was the strongest predictor of VTE, followed by clinical factors (BMI, ABO alleles, age) and expression of 494 genes, including STS, FAM13A, GPRIN1, FLVCR2, FAM177B, and several long non-coding RNAs not previously linked to thrombosis. Known thrombosis-associated genes such as UQCRC2 and PRKRA were also identified. Significant enrichment was observed for cardiomyopathic KEGG pathways and renal HPA terms. Similarity-based risk score construction improved classification, with 74% of VTE cases and 23% of controls assigned to the risk zone.CONCLUSIONS: Multivariate integration via machine learning enhances VTE risk stratification, identifying novel transcriptomic signatures and lncRNA biomarkers that offer new strategies for VTE personalized prevention.

KW - RNA-seq

KW - Machine Learning

KW - Idiopathic Venous Thromboembolism

KW - risk score

UR - https://www.scopus.com/pages/publications/105041853270

U2 - 10.1016/j.jtha.2026.05.015

DO - 10.1016/j.jtha.2026.05.015

M3 - Article

C2 - 42167585

SN - 1538-7933

JO - Journal of Thrombosis and Haemostasis

JF - Journal of Thrombosis and Haemostasis

ER -