Abstract
ABSTRACTN6-methyladenine is an epigenetic modification that plays a significant role in various cellular processes. Genome-wide monitoring of methylation sites is conducive to understanding the biological function of methylation. Due to the limitations of traditional dry and wet experiments, a series of machine learning and deep learning methods have been developed to detect methylation sites, but their detection species are single or performance is poor. First of all, we conducted sufficient experiments on the widely studied rice datasets, and compared with the previous research, we have greatly improved in various indicators on the two rice datasets. Then we used the models trained on the rice dataset to fine-tune training in half of the other 11 datasets and predict the other half of the independent datasets. Then we used 11 trained models to test 11 species respectively. It was found that ACNN-6mA could obtain higher AUC, ACC and MCC whether cross-species prediction or independent verification set prediction. ACNN-6mA model and code for follow-up researchers is provided as an open-source tool available athttps://github.com/jrebai/ACNN-6mA.
Publisher
Cold Spring Harbor Laboratory