CAP-m7G: A capsule network-based framework for specific RNA N7-methylguanosine site identification using image encoding and reconstruction layers

Abstract

N7-methylguanosine (m7G) modifications play a pivotal role in RNA stability, mRNA export, and protein translation. They are closely associated with ribosome function and the regulation of gene expression. Dysregulation of m7G has been implicated in various diseases, including cancers and neurodegenerative disorders, where the loss of m7G can lead to genomic instability and uncontrolled cell proliferation. Accurate identification of m7G sites is thus essential for elucidating these mechanisms. Due to the high cost of experimentally validating m7G sites, several artificial intelligence models have been developed to predict these sites. However, the performance of these models is not yet optimal, and a user-friendly web server is still needed. To address these issues, we developed CAP-m7G, an innovative model that integrates Chaos Game Representation, Capsule Networks, and reconstruction layers. CAP-m7G achieved an accuracy of 96.63%, a specificity of 95.07%, and a Matthews correlation coefficient (MCC) of 0.933 on independent test data. Our results demonstrate that the integration of Chaos Game Representation with Capsule Network can effectively capture the crucial sequence information associated with m7G sites. The web server can be accessed at https://awi.cuhk.edu.cn/~biosequence/CAP-m7G/index.php.