Research on a Deep Hopfield Neural Network for Class Prediction of Breast Cancer Gene Data

Cancer has always been an unresolved problem in the medical field. Until now, researchers have proposed a variety of methods for unsupervised classification of gene subtypes to determine the genetic characteristics associated with the identified subtypes. The traditional Hopfield neural network has the advantages of fast and simple structure, but its robustness is poor. Therefore, on the basis of the previous research, this paper focused on the new patient’s class prediction problem, making appropriate improvements of the Hopfield network, and proposed a deep Hopfield neural network. The data set for model training and testing is the public gene expression data set obtained from the http://www.broadinstitute.org/cgi-bin/cancer/datasets.cgi. This paper used the R Programming Language to preprocess the data and draw the heat map. By analyzing the performance of the Hamming distance and entropy of the data set in the Hopfield neural network, we found that the memory depth of the network has a significant influence on the classification results. As a result, our method is proved to be effective and more accurate using the same data which NTP and Hope4Genes originally applied.