iHNHC-RsFPN:Prediction of Human Non-histone Crotonylation Sites Based on Multi-feature and Feature Pyramid Networks
10.13865/j.cnki.cjbmb.2025.06.1164
- VernacularTitle:iHNHC-RsFPN:基于多特征和特征金字塔网络预测人类非组蛋白巴豆酰化位点
- Author:
Xin WEI
1
;
Si-Qin HU
;
Jian TU
;
Muhammad Akmal REMLI
Author Information
1. Institute for Artificial Intelligence and Big Data,Universiti Malaysia Kelantan,Kota Bharu,Malaysia;江西服装学院商学院数理统计教研室,南昌 330201;Faculty of Data Science and Computing,Universiti Malaysia Kelantan,Kota Bharu,Malaysia
- Publication Type:Journal Article
- Keywords:
human non-histone proteins;
crotonylation site;
deep learning;
feature extraction;
ensemble learning;
residual neural network(RsFPN)
- From:
Chinese Journal of Biochemistry and Molecular Biology
2025;41(10):1541-1551
- CountryChina
- Language:Chinese
-
Abstract:
Human non-histone lysine crotonylation plays crucial roles in biological activities.However,traditional wet-lab experiments are time-consuming and labor-intensive,making computational prediction methods increasingly popular in recent years.Despite the biological importance of lysine crotonylation,there are relatively few studies on human non-histone proteins.In this study,we developed an ensemble deep learning predictor named iHNHC-RsFPN by constructing a Residual Pyramid Network(RsFPN).First,three feature extraction methods were employed to encode sequence samples.Next,weak classifi-ers based on RsFPN were individually trained for different feature types.Finally,these weak classifiers were integrated to build a robust final predictor.Independent test results demonstrated that iHNHC-RsF-PN achieved outstanding performance across four key metrics:sensitivity(Sn=0.8580),specificity(Sp=0.7463),accuracy(Acc=0.7798),and Matthews correlation coefficient(MCC=0.5586).Comparative experiments revealed that iHNHC-RsFPN significantly improved prediction accuracy for hu-man non-histone crotonylation sites over existing methods.Additionally,we established a user-friendly web server(http://www.lzzzlab.top/ihnc/)that provides straightforward prediction services without complex calculations,facilitating further research for experts in related fields.
