The Analysis Of Machine Learning-Assisted Mutation Evolution Of Creatinase

Creatinase is often used to measure creatinine in clinical practice to evaluate renal function.Meanwhile,Creatinase also an important role in industrial production,but the original enzyme is normal temperature enzyme,and its natural evolution speed is slow.Therefore,there is an important topic of current research about how to imitate the key steps of natural evolution in the laboratory to design sequence mutation of protease,so that to improve its thermostability.The diversity of physical space structure directly influences the thermostability of protein,and protein sequence only determines its structure space.Therefore,we start from the sequence and hope to improve its thermostability by generating new sequences through mutations at some sites.Based on a certain understanding of creatinase biological characteristic,18 sites were found in the sequence for mutation,and each site needs to select mutation or not mutation,so there are many selections.It is difficult to evaluate its thermostability by artificial experiment because of the huge amount of dataset.The newly proposed protein mutation evolution technique assisted by artificial intelligence can help us solve the above screening bottleneck.In this paper,artificial intelligence technology is used to establish the relationship model between sequence and thermostability after coding labeled sequences,we select appropriate models and obtain some combination models to predict the thermostability of unlabeled mutant sequences.Finally,mutants with significantly improved thermostability were found through the predicted values and experimental verification.The innovation of this article is not only lies in the data observed by our project partner.We help them find the best mutant,and we provide higher thermostability creatine mutants.It also lies in the way how the final model is chosen,this paper used the method of model combination.we have selected some models with higher scores in testing set as the candidate model,through the performance of the real value to calculate the weight.Then we had some combination model and improtant model.These models improve prediction accuracy.