基于注意力机制的肝癌基因表达识别

打开文本图片集

中图分类号：TP183；TP391 文献标识码：A 文章编号：2096-4706（2025）15-0043-05

Abstract： Hepatocellular carcinoma is a type cancer with high morbidity mortality in ， which brings a heavy diseaseburden tothe society.Inrecent years，withthein-depthdevelopmentaplicationDeepLearning technologysome algorithms basedonDepLeamingto predict gene expressionusing generegulatory factors have beenproposed.However， there is stillmuchroomforimprovement intheaccuracythealgorithmforpredictingtheup-regulationdown-regulationgene expressionbyhistone modificationinhepatocelularcarcinoma.Inthispaper，aDepLearningalgorithmmodelACRChrome is proposed，whichcanpredictthechangesgeneexpressionbyusingthediferencehistone modificationsignalsbetween hepatocellularcareomacellinesndnoallivercelli.ItusesSelf-AtentonMechasm，To-DimesioalConvolioal Neural Network （2D-CNN）residualconnection toobtain feature importance fromdata，capture data depth features aleviate thevanishinggradient problem.TheresultsshowthatteaverageAUC，averageprecision，averagesensitivityverage specificityaverageF-scoreaverage accuacyCRChromerepeatedtraining for5timesare0.885，0.833，0.8，0.71， 0.843 0.803，respectively.Afterrepeated trainingfor15 times，the AUC index ACRChrome is significantly improved compared withthe same type advanced model，indicating that the model can effectively predict gene expression.

KeyWords：hepatocelllarcarcinoma;histone modification;Self-Attention Mechanism;2D-CNN;residualconection

0 引言

肝癌是一种发生在肝脏的恶性肿瘤，可分为原发性和继发性两大类。（剩余9379字）