基于多模态特征融合的米氏常数预测模型

打开文本图片集

中图分类号：R318.04 文献标志码：A 文章编号：2095-2945(2025)30-0028-05

Abstract:Michaellanconstantisakeyparameterinenzyekineticsandreflectstheafinitybetweenenzymeandsubstrate. Accuratelypredictingthisvalueiscrucialforreshapingthemetabolicnetwork，whichprovidesthebasisfordesigningand optimizingcellfactories.However，mostcurentdeeplearningmodelsdonotfullyminethestructuralinformationofproteinsand fail tomakefulluseofthesequenceinformationandstructuralinformationofenzymesandsubstrates,thusaccuratelypredicting Michaelisconstantsisstillchalenging.Tsolvethisproblem,anewdeeplearningmodelisproposedthatintegratesmultiodal information,includingproteinsequences，proteinmaps，substrate SMILESstringsandmolecularmaps,andusesaconvolutional atentionmoduleforfeaturefusion.Theresultsshowthatthemodelcombiningmultimodalinformationissuperiortoexisting methods and isa powerful tool for predicting the kinetic parameters of enzymatic reactions.

Keywords:biocatalysis;biological processes;synthetic biology；deep learning;neural network;Michaelisconstant

酶是精确控制代谢途径通量的关键催化剂，酶动力学参数在代谢网络的重构中发挥着核心作用。（剩余10706字）