MATLAB Applications For Forecasting The1,4-Dihydropyridines’ Quantitative Structure Activity Relationship

In this paper, the research profile of the medicine screening forecast in MATLAB issummarized. These numerical prediction methods can be divided into two categories: theRegression Equation and Neural Network. The forecast of Regression Equation wascalculated from the known data, and obtained the unknown data from the fit line trend ofthe Regression Equation. In order to get the unknown data, the known data set is mappedfrom to the unknown ones according to certain learnable principles in essence byArtificial Neural Networks. First, In the MATLAB the theories of Neural Network andRegression Equation is analysed. Then put forward the forecast models based on eachmethod. Detailed the process of model founding, including the molecule of drugsdescriptor’s choice, training and testing data preparation, etc. The article take the forecastof1,4-dihydropyridines’melting point and50%effective concentration for blockingCa2+channel(EC50) as example, two methods have been used in this project: one is touse the Regression Equation in establishing the model, the other one is to switch the setof the sample into the region [-1,1]. And then set up the model of Neural Network.While in the latter method, we use the two commonly function in neural network to formthe modeling respectively. And test different hidden layer nodes. Set up modeling andexperiments in these two ways by contrast, the results show that: although neuralnetwork prediction model is more cumbersome in the early process of data processing aswell as choosing the middle hidden layer nodes, A well trained Artificial Neural Networkis better in forecasting accuracy rate than the Regression Equation method. Use ArtificialNeural Network to forecast the1,4-dihydropyridines drugs’ EC50is a better method.The studied models in this research provide meaningful reference for similar drugs suchas1,4-dihydropyridines’ screening and designing.