›› 2020, Vol. 11 ›› Issue (3): 151-155.

Previous Articles     Next Articles

Machine learning based human oral epithelium active enhancers predicts a pathogenic variant associated with orofacial cleft near IRF6

  

  • Received:2020-07-19 Revised:2020-08-26 Online:2020-09-25 Published:2020-09-30
  • Contact: Huan Liu E-mail:liu.huan@whu.edu.cn

Abstract: Objective: This study aimed to validate whether machine learning based on craniofacial-specific oral epithelium enhancers could be used to non-coding variants associated with orofacial cleft. Methods: Anti H3K27Ac ChIP-seq was performed in human immortalized oral epithelial cells (HIOEC), which were integrated published craniofacial super-enhancers. The sequences of these enhancers were used as training set for gapped k-mer SVM (gkm SVM) machine learning, which could summarize all the DNA sequence features. Delta SVM scores were employed to validate the effects of non-coding variants near IRF6, and dual luciferase assays were used for biological validation. Results: Published orofacial cleft-associated SNPs were more enriched in oral epithelium-specific enhancers. gkm SVM showed 350dupA mutation could significantly decrease enhancer activity, which was then validated using dual luciferase assays in HIOEC cells. Conclusions: The classifiers based on oral epithelium-specific enhancers are useful in nominating functional SNPs identified in orofacial clefting genome wide association studies.

Key words: cleft lip and palate, functional genome, machine learning, IRF6