Genome wide prediction of monoallelic gene expression from human epigenetic data

Monoallelic expression is a process where transcription occurs from only one allele of a diploid organism. X-Chromosome inactivation in females and genomic imprinting are relatively well studied forms of monoallelic expression. Apart from them, about 15\% of autosomal genes show random monoallelic expression (aRME). aRME is widely observed in gene families of the immune and nervous system like olfactory receptor genes, immunoglobulins, protocadherins, etc. However, recent genome wide studies suggest aRME is observed in individual autosomal genes like Albumin, Interleukin, etc. Among them some genes show mitotically stable and cell-cell dynamic forms of aRME, inclusive of genes involved in various human diseases. Because of the stochastic nature of aRME, these genes are difficult to characterize and so relatively under studied. Nag and coworkers found that H3K27me3 and H3K36me3 are the chromatin signature of monoallelic expression. In this work, we tried sampling methods, and explored various histone marks and their different enrichment regions as features to evaluate various classificiation methods. Finally, we chose a random forest classifer using gene body signals of H3K27me3, H3K36me3, H3K27ac, H3K9me3, H3K4me1, and H3K4me3 as features to predict monoallelic expression, with a F-measure of 0.90. Using this classifer, we also predicted monoallelic genes across various human tissues and cell lines by obtaining epigenetic data from various consortia. Genes involved in single and multicelluar organismal and developmental process were found enriched among the predictions. We hope our classifier and the predictions will help in improving the understanding of monoallelic gene expression.