Workshop TFDA Comp-Bio

CTpredX: Enhancing Missense Variant Pathogenicity Prediction in Childhood Cancer Predisposition Genes

Not scheduled

Complesso dei SS. Marcellino e Festo

Description

Effective clinical genome interpretation relies on accurately distinguishing between benign and pathogenic rare variants. Current genome-wide trained variant prioritization tools often lack precision and may overlook key parameters defining gene-disease relationships. We hypothesized that developing predictors based on curated, disease-specific gene sets could significantly enhance the performance of these tools.

We developed CTpredX, a machine learning-based predictor specifically tailored for a curated set of childhood cancer predisposition genes (CCPGs). CTpredX was trained through an Extreme Gradient Boosting algorithm using ClinVar annotations as a ground truth, integrating multiple variant annotations, constraint scores, and existing pathogenicity predictors.

CTpredX outperformed existing genome-wide tools in distinguishing pathogenic from benign variants within CCPGs, achieving an AUC-ROC=0.98 in the testing dataset. Benchmarking against established prediction tools also demonstrated superior performance in terms of accuracy in distinguishing pathogenic from benign variants at high confidence levels. Additionally, CTpredX re-classified 56% of ClinVar's variants of uncertain significance (VUS), with 37.8% of them re-classified as benign and 18.3% as pathogenic.

CTpredX provides a robust tool for pathogenicity prediction of missense variants in childhood cancer, supporting the feasibility of disease-specific variant classifiers. A user-friendly R-shiny interface facilitates broader use in clinical and research settings.

Speaker recent publications:

• Bonfiglio F, Lasorsa VA, Aievola V, Cantalupo S, Morini M, Ardito M, Conte M,
  Fragola M, Eva A, Corrias MV, Iolascon A, Capasso M. Exploring the role of HLA
  variants in neuroblastoma susceptibility through whole exome sequencing. HLA.
  2024 May;103(5):e15515. doi: 10.1111/tan.15515. PMID: 38747019.

• Torices L, Zamfir-Taranu A, Esteban-Blanco C, Bozzarelli I, Bonfiglio F,
  D'Amato M. Human CAZyme genes polymorphism and risk of IBS: a population-based
  study. Gut. 2024 Jul 5:gutjnl-2024-333056. doi: 10.1136/gutjnl-2024-333056. Epub
  ahead of print. PMID: 38969488.

• Bonfiglio F, Lasorsa VA, Cantalupo S, D'Alterio G, Aievola V, Boccia A,
  Ardito M, Furini S, Renieri A, Morini M, Stainczyk S, Westermann F, Paolella G,
  Eva A, Iolascon A, Capasso M. Inherited rare variants in homologous
  recombination and neurodevelopmental genes are associated with increased risk of
  neuroblastoma. EBioMedicine. 2023 Jan;87:104395. doi:
  10.1016/j.ebiom.2022.104395. Epub 2022 Dec 6. PMID: 36493725; PMCID: PMC9732128.

• Bonfiglio F, Liu X, Smillie C, Pandit A, Kurilshikov A, Bacigalupe R, Zheng
  T, Nim H, Garcia-Etxebarria K, Bujanda L, Andreasson A, Agreus L, Walter S,
  Abecasis G, Eijsbouts C, Jostins L, Parkes M, Hughes DA, Timpson N, Raes J,
  Franke A, Kennedy NA, Regev A, Zhernakova A, Simren M, Camilleri M, D'Amato M.
  GWAS of stool frequency provides insights into gastrointestinal motility and
  irritable bowel syndrome. Cell Genom. 2021 Dec 8;1(3):None. doi:
  10.1016/j.xgen.2021.100069. PMID: 34957435; PMCID: PMC8654685.