Predicting enhancer activity and variant impact using gkm-SVM.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Understanding the impact of genetic variants on gene regulation is crucial for deciphering human disease.
• Computational models are increasingly used to predict the functional consequences of genetic variations.

Purpose of the Study:

• To evaluate the performance of a gapped-kmer support vector machine (gkm-SVM) model in predicting regulatory variant impact.
• To compare gkm-SVM with other models in the Critical Assessment of Genome Interpretation eQTL challenge.
• To assess the utility of different genomic data types for training predictive models.

Main Methods:

• Utilized a discriminative gapped-kmer SVM (gkm-SVM) model trained on genome-wide chromatin accessibility data.
• Compared gkm-SVM predictions with experimental data from massively parallel reporter assays (MPRA) in lymphoblasts and K562 cells.
• Evaluated the predictive power of DNase-I hypersensitive sites (DHS) and assay for transposase-accessible chromatin using sequencing (ATAC-seq) data, with and without histone mark information.

Main Results:

• gkm-SVM demonstrated high accuracy in predicting eQTLs, outperforming models trained on MPRA data.
• gkm-SVM proved to be a reliable predictor of gene expression and variant impact across different cell types and tissues.
• Both DHS and ATAC-seq data were equally effective for training gkm-SVM; DHS regions with H3K27Ac and H3K4me1 histone marks showed enhanced predictive capability.

Conclusions:

• The gkm-SVM model is a robust and accurate tool for predicting the impact of regulatory variants on gene expression.
• Chromatin accessibility data, particularly when combined with specific histone marks, provides a strong foundation for training predictive genomic models.
• This approach holds promise for advancing the interpretation of genetic variations in the context of human disease.