Optimal Mathematical Modeling of Drug Therapy Based on Gray Correlation Analysis and Neural Network

In the last decade, the prevalence and mortality of breast cancer are incremental, which critically intimidates women's health all over the world. Accordingly, to research and analyse the alternative drugs for curing breast cancer is essential. Peculiarly, the quantity of breast cancer patients accompanied by ER. α is giant, and mechanism of the disease is intricate. First of all, this paper carries on the data preprocessing, removes 225 molecular descriptors whose data are all zero, and retains 504 molecular descriptor data as the correlation analysis object of the grey correlation method. Based on the Matlab software, the grey relational analysis model is established. The correlation degree weight of each molecular descriptor is obtained, sorted from large to small, to judge the important extent of the infection of each molecular descriptor on biological activity. Finally, the first 20 molecular descriptors with the most striking effect on biological activity were filtrated. The Spearman correlation analysis of 20 selected molecular descriptors is carried out using Matlab software to test the independence of the index. The color in the distribution map of the Spearman correlation coefficient is basically very light, and most of the results of the p-value test are less than 0.05. It shows significant differences among molecular descriptors, which proves that the selected molecular descriptors have strong independence. It also shows that the grey relational analysis model is very reliable. Definitively, the quantitative prediction model of biological activity on account of neural network and multiple linear regression are set up. The parameters of the quantitative prediction model based around neural network are set as follows: there are fifteen neurons in the hidden layer, and the magnitude of iterations is 1000, and also the disciplinal goal is 0.0000001. On account of the quantitative prediction model set by parameters, the unitary BP neural network structure hidden layer is devoted to inspect the model. Finally, by comparing the predicted values of the two models, it is found that the predicted value distribution of the multiple linear regression quantitative prediction model is stable.

I. Introduction

In the last decade, the prevalence and mortality of breast cancer in China have incremented in developed and developing countries. Breast cancer constitutes a nasty intimidation to universal women's health. According to the results, the incidence rate of breast cancer is closely related to the estrogen receptor. The experimental results in mice showed that estrogen receptor a Subtype (estrogen-receptor alpha, ER) a). It plays a key role in breast development. Therefore, er a, Being used as a key target for breast cancer treatment. It can target ER a Compounds antagonistic to activity will be candidates for breast cancer treatment. However, as a candidate drug for breast cancer treatment, it also has eminent pharmacokinetic properties and safety. It is referred to as ADMET (Absorption absorption, Distribution distribution, metabolism, Excretion excretion, toxicity), and its properties are optimized.