CAJ | 학술논문

【目的】pH值是影响酶催化效率的关键参数，通常需要通过实验方式才能确定生物酶的最适 pH 值，而该方式要消耗较多的人力、物力和时间。因此，有必要发展一种利用酶的简单结构信息即可预测其最适 pH 值的方法。【方法】以20-1前馈反向传播的神经网络为模型，完成535种氨基酸属性对α-淀粉酶 pH 值的拟合。同时，将α-淀粉酶 Amy7C及其54个突变体的数据分为2组，用35个酶作为训练组进行拟合，20个酶作为验证组进行检验，并对不同层次及神经元个数的模型进行比较。【结果】109个氨基酸属性可实现20-1神经网络模型收敛，表明这些氨基酸属性可用于预测α-淀粉酶的最适 pH值，但是不同氨基酸属性预测 pH 值的效果差别较大，只有部分指标预测 pH值的效果较好。多模型的分析结果显示，不同模型对训练组R值的结果具有显著性差异，而对训练组P值、验证组R值和验证组P 值结果无显著性差异。【结论】氨基酸分布概率等属性可用于预测α-淀粉酶的最适 pH值。20-1神经网络模型是预测α-淀粉酶最适 pH值相对理想的模型。
【목적】pH치시영향매최화효솔적관건삼수，통상수요통과실험방식재능학정생물매적최괄 pH 치，이해방식요소모교다적인력、물력화시간。인차，유필요발전일충이용매적간단결구신식즉가예측기최괄 pH 치적방법。【방법】이20-1전궤반향전파적신경망락위모형，완성535충안기산속성대α-정분매 pH 치적의합。동시，장α-정분매 Amy7C급기54개돌변체적수거분위2조，용35개매작위훈련조진행의합，20개매작위험증조진행검험，병대불동층차급신경원개수적모형진행비교。【결과】109개안기산속성가실현20-1신경망락모형수렴，표명저사안기산속성가용우예측α-정분매적최괄 pH치，단시불동안기산속성예측 pH 치적효과차별교대，지유부분지표예측 pH치적효과교호。다모형적분석결과현시，불동모형대훈련조R치적결과구유현저성차이，이대훈련조P치、험증조R치화험증조P 치결과무현저성차이。【결론】안기산분포개솔등속성가용우예측α-정분매적최괄 pH치。20-1신경망락모형시예측α-정분매최괄 pH치상대이상적모형。
Objective]pH is an important parameter in enzymatic reaction,and its determination is often through experimental path,which is generally costly and time-consuming.So,it is nec-essary to develop methods that can use as simple as possible information to predict pH opti-mum for enzyme.[Methods]20-1 feedforward backpropagation neural network was used to screen 535 amino acids properties as predictors to predict the optimal pH ofα-amylase Amy7C and its 54 mutant,which were divided as two groups.35 of them served as training group for fitting,and the other 20 were treated as valida-tion.The models for different structures and neuron numbers were also compared.[Results]109 amino acid properties,which converged dur-ing fitting in the 20-1 neural network model, could be used to predict optimal pH.Different a-mino acid properties presented different predic-ting effect,and some of them revealed better prediction for optimal pH.The multi-model results showed that there was significant difference between R values in training groups,but there was no significant difference between P values in training groups,as well as R and P values in validation groups.[Conclusion]The distribution probability and some amino acid properties could be used to predict optimal pH ofα-amylase, for which 20-1 feedforward backpropagation neural network was the relative ideal model.