Abstract

Epigenetic mechanisms play a crucial role in regulating the expression of genes affecting the development, growth, and functioning of organisms, ensuring the activation or repression of specific genes at the appropriate times and in the correct cells. This functionality enables organisms to adapt to internal and external stimuli, uphold homeostasis, and execute diverse biological processes. One such mechanism modulating the expression of genes involves modification of histone proteins. Consequently, there is a need for identifying and comprehensively understanding various histone modifications and their effect on gene expression. The laboratory-based identification process entails the examination of histone modifications (HMs) to analyze the chemical alterations of histone proteins associated with DNA. On the other hand, technological advancements empower AI to address these challenges by discerning protein chemical alterations at HMs associated with DNA, facilitating accelerated comprehension of gene expression with maintained precision, result accuracy, and substantial cost reduction. In this paper, we introduce TransformerChrome, a computational model-based on the transformer neural network architecture, designed to take HMs from gene sites as input and predict gene expression as output. The developed TransformerChrome model has been trained and tested across 56 distinct cell types, using five core HMs. The presented model has been compared against state-of-the-art HM based gene expression predictive models across benchmark datasets in humans. The model is evaluated in terms of area under the receiver operating characteristic curve. The results show superior performance, especially on cell types for which other models show significantly lower performance.

Article ID: 2024S5

Month: May

Year: 2024

Address: Online

Venue: The 37th Canadian Conference on Artificial Intelligence

Publisher: Canadian Artificial Intelligence Association

URL: https://caiac.pubpub.org/pub/x9w6ew2i