An Effective Algorithm Based on Sequence and Property Information for N4-methylcytosine Identification in Multiple Species

Abstract: N4-methylcytosine (4mC) is one of the most important epigenetic modifications, which plays a significant role in biological progress and helps explain biological functions. Although biological experiments can identify potential 4mC sites, they are limited due to the experimental environment and labor-intensive process. Therefore, it is crucial to construct a computational model to identify the 4mC sites. Some computational methods have been proposed to identify the 4mC sites, but some problems should not be ignored, such as those presented as follows: (1) a more accurate algorithm is required to improve the prediction, especially for Matthew’s correlation coefficient (MCC); (2) easier method is needed for clinical research to design medicine or treat disease. Considering these aspects, an effective algorithm using comprehensible encoding in multiple species was proposed in this study. Since nucleotide arrangement and its property information could reflect the sequence structure and function, several feature vectors have been developed based on nucleotide energy information, trinucleotide energy information, and nucleotide chemical property information. Besides, feature effect has been analyzed to select the optimal feature vectors for multiple species. Finally, the optimal feature vectors were inputted into the CatBoost algorithm to construct the identification model. The evaluation results showed that our study obtained the highest MCC, i.e., 2.5%~11.1%, 1.4%~17.8%, 1.1%~7.6%, and 2.3%~18.0% higher than previous models for the A. thaliana, C. elegans, D. melanogaster, and E. coli datasets, respectively. These satisfactory results reflect that the proposed method is available to identify 4mC sites in multiple species, especially for MCC. It could provide a reasonable supplement for biological research.