多特征结合机器学习算法识别蛋白质与组胺的结合位点OA

Histamine,a bioactive molecule crucial in various diseases,plays a significant role in disease pathogenesis.An in-depth study of protein-histamine interactions helps to understand their binding mechanisms.This study combined multiple features with various machine learning algorithms and ensemble methods to predict proteins-histamine binding sites.The results showed that the gradient boosting(GB)model and average flexibility index(AF)feature within the machine learning algorithms yielded the best pre-diction results,achieving an accuracy(A)of 82.1%,a Matthews correlation coefficient(M)of 68.2%,and a sensitivity(S)of 65.5%.The multi-feature fusion and ensemble algorithm processing was applied to the codon usage frequency feature,which had poor prediction performance.The results indicated that the performance of the improved model was significantly enhanced.In fea-ture fusion,the combination of codon frequency and average flexibility index resulted in all metrics exceeding 60.0%.When the co-don frequency was used as a single feature in the ensemble algorithm prediction model,the accuracy(A)increased by 19.0%,and the Matthews correlation coefficient(M)increased by 66.1%compared to the optimal gradient boosting(GB)model.The result demonstrated that feature fusion and ensemble algorithms effectively predicted histamine binding sites.