Prediction of lysine crotonylation sites iKcr-DRC: Prediction of lysine crotonylation sites in proteins based on a novel attention module and DenseNet

Xin Wei^1,2Siqin Hu³Jian Tu²Muhammad Akmal Remli^1,4*

• ¹Insitute for Artificial Intelligence and Big Data, Universiti Malaysia Kelantan, Kota Bharu, Kelantan Darul Naim, Malaysia
• ²Business School, Jiangxi Institute of Fashion Technology, Nanchang, Jiangxi, China
• ³School of Mega Data, Jiangxi Institute of Fashion Technology, Nanchang, Jiangxi, China
• ⁴Faculty of Data Science and Computing, Universiti Malaysia Kelantan, Kota Bharu, Kelantan Darul Naim, Malaysia

Lysine crotonylation (Kcr) is a novel post-translational modification of proteins that has attracted wide attention in recent years, mainly occurring on protein lysine residues. It plays a pivotal role in gene expression regulation, cellular metabolism, and a variety of life activities. Kcr has an important regulatory role in major diseases such as tumors and cancer. Therefore, accurate prediction of the Kcr site is essential for understanding the normal functioning of an organism. We propose a deep learning computational model called iKcr-DRC for accurate prediction of Kcr sites. In this model, we employ densely connected convolutional networks (DenseNet) as the core network structure to extract advanced local feature information. In addition to this, we have made innovative improvements to the channel attention mechanism. We have designed a short-circuit connection structure within the channel attention mechanism, which gives it residual properties. The experimental results show that the iKcr-DRC model achieves 90.30%, 78.35%, 84.33% and 69.15% for sensitivity, specificity, accuracy, and Matthew’s correlation coefficients, respectively. The iKcr-DRC model outperforms other existing predictors in predicting lysine crotonylation sites, and this study provides new ideas and methods for the application of deep learning in bioinformatics.