Based On Deep Learning For Drug Activity Research

With the high speed development of the world economy,the efficacy study based on the biologically active molecules has been improved.At present,the combination of pharmacological analysis,high-throughput screening technology,mathematical statistics and other technologies has opened a new journey for the study of intrinsic properties of drug activity.However,because of the high latitude,high complexity of drug molecules,converting the various types of technology to practical operation is still very difficult and it becomes a bottleneck in the scientific research process.Therefore,it is an urgent task for researchers to determine the active molecules in a large amount of pharmacological data.In order to find the way quickly to solve the problem,researchers use the computer as a tool for drug discovery.To a large extent,the use of computer has played a role in promoting the work of scientific research workers.However,for the activity of drug molecules,most of the researchers use only one type of the existing calculation methods in the study,this method constraints the detection range of drug reactive molecules in a certain extent,and it's also unfavorable for finding drugs timely.In the actual process of testing,it is easier to obtain the no label data than the label data.Therefore,according to the sample properties,this text uses two types of computer aided algorithm to research drug reactive molecules,respectively shallow machine learning and deep machine learning.shallow machine learning is divided into supervised algorithm and semi-supervised algorithm.deep machine learning is unsupervised algorithm.In supervised algorithm,support vector machine(SVM)and artificial neural network(ANN)are more common.In semi-supervised algorithm,semi-supervised support vector machine(S4VM)and cost security semi-supervised support vector machine(CS4VM)are more representative.In unsupervised algorithm,stacked auto encoder(SAE)and deep belief network(DBN)is more outstanding.For the purpose of the study,this article will allocate these six methods appropriately to explore the three types of drugs active molecule(PLK1 PBD,SMAD3,IL-1B).Because the structure of drug active molecules is complex,choosing Chemometric software MOE to have a precise calculation,for getting molecular descriptors of 2D and 3D,through molecular information pretreatment and use those two methods we above-mentioned to recognize the drug molecules,the experimental results show that under the same conditions,unsupervised algorithm is easier to extract deep information of active molecules and compared with other algorithms,the accuracy,sensitivity,specificity and accuracy etc indicators are have obvious advantage.