| In recent years,with the increasing number of cancer patients,the mortality rate of cancer has also risen.For most cancer patients,traditional physical and chemical treatments are expensive and inefficient.In addition,some anti-cancer drugs can damage normal cells while killing cancer cells,and patients who take these drugs for a long time can make cancer cells resistant.Therefore,there is an urgent need to develop new and more effective drugs.As a safe and reliable therapeutic drug candidate with high specificity and selectivity,polypeptide has been favored by researchers in recent years.With the increase of peptide drug data and the rapid development of machine learning algorithms,mining the characteristics of peptide drugs for new drug design has become a research hotspot.Although researchers have achieved quite a lot of achievements in the identification and prediction of peptides based on machine learning algorithms,there is still room for further research in feature extraction,classification methods and peptide design.Based on the theory of machine learning,this dissertation studies the characteristic relationship between amino acids and amino acids,the identification of peptide drugs,and the problems related to the generation of peptide drugs.The specific research work is summarized as follows:1.Study on the characteristics of amino acids-amino acids in anticancer peptide drugs.When studying the characteristic relationship between amino acids and amino acids in polypeptides,it is far from enough to only consider sequence information.In view of this,this paper not only considers the relationship between the adjacent dipeptide and the dipeptide with space,but also considers the information of the secondary structure,that is to judge whether the amino acid and the amino acid are in the same secondary structure,and then design different scoring rules are designed,and the prediction accuracy and search time of grid search and random search methods are compared.The results show that after sacrificing a little prediction accuracy,the search time of random search is nearly 2000 times less than that of grid search.Scoring rules and random search are combined to find the best parameters.In order to further explore the relationship between polypeptide sequences,multiple sequence alignments are used to find sequences in the same family,to obtain the accuracy of multiple sequence alignments and the sequence conservation of each position.To aid the experiment,this paper also proposes a method of calculating probabilities to find conservative positions.2.Identification of anticancer peptide drugs based on machine learning.At present,good research results have been achieved in the study of peptide sequence description methods,but the research on the structure of peptides is relatively rare.In this study,anti-cancer peptides and anti-hypertensive peptides were taken as the research objects,and the information of the primary,secondary and tertiary structures of peptides was considered respectively.And a new method for describing peptide drugs is proposed,which uses topological attribute values(degree,proximity centrality,betweenness centrality)in complex networks to describe peptide drugs from various levels.Then the model is constructed based on three algorithms of support vector machine,K-nearest neighbor and random forest.In order to make the method proposed in this paper convincing,the proposed method is compared with two methods of others.It can be seen from the results that compared with the existing methods,the model can predict anti-cancer polypeptide drugs and anti-hypertensive polypeptide drugs well.Furthermore,the generalization ability of the model is well validated by constructing 3 independent test datasets.In order to obtain the salient features that distinguish anti-cancer peptide drugs and anti-hypertensive peptide drugs,this paper uses a feature selection algorithm based on support vector machine recursive feature elimination,and according to weight ranking,the important features obtained are Trp,Ala,Asn,Val,Glu,Ile,Lys,Leu,Arg,Tyr;the three types in the secondary structure are α-helix,turn and coil;the important forces obtained in the tertiary structure are hydrogen bond and van der Waals force.To illustrate that the features chosen in this paper are reasonable,the length of the two peptide drugs,the content of amino acids in the primary sequence,the binding tightness of amino acids to the eight types of secondary structures,and the strong and weak interactions in the tertiary structure were analyzed.The four aspects of force distribution are analyzed and discussed,and it is found that they are consistent with the selected important characteristics.In order to facilitate the research of other scholars,this paper provides a free online prediction platform based on the Django framework.3.Generation of anticancer peptide drugs based on long short-term memory network.As a new type of drug for the treatment of cancer,anti-cancer polypeptide drugs are expected to become the best treatment for various tumor diseases because of their small side effects.With the passage of time,the drug resistance of anti-cancer drugs continues to rise,and the generation of new anti-cancer peptide drugs has become an urgent need.This paper firstly analyzes the research status of methods for generating new peptide sequences,and focuses on the principle of generating new peptide drugs based on long short-term memory network algorithm.Judging the performance of the resulting peptide drug is incomplete.Therefore,relevant evaluation indicators such as bilingual evaluation studies are added to calculate the similarity between the generated sequence and the original sequence,so as to better evaluate the performance of the generated peptide sequence.At the same time,the sampled new and original sequences are input into the random forest algorithm for prediction,and the results show that the generated sequences are reliable.In summary,this paper firstly takes dipeptide as the research object,conducts characteristic research based on primary structure and secondary structure respectively,and deeply understands the sequence conservation of species based on multiple sequence alignment algorithm,so as to identify the most critical sequences for peptide function.Then,a new method for describing the characteristics of polypeptide drugs was proposed,which combined the primary structure,secondary structure and tertiary structure information of peptides and used them for the identification of peptide drugs.The experimental results showed that the proposed method in this paper The method can better identify peptide drugs,and the feature selection algorithm is helpful to mine the key features of peptide drugs and assist subsequent scholars’ research;finally,new peptide drugs are generated based on long-term and short-term memory network algorithm,and the sampled samples are analyzed in function analysis,Its performance was evaluated on the relevant evaluation indicators such as physical and chemical properties and bilingual evaluation research.The innovation of this paper is mainly reflected in the system perspective,using network science theory to extract peptide features,accurately identify the two types of drugs,and further excavate the key features of the two types of peptide drugs,providing a theoretical basis for the analysis and design of new peptide drugs. |
