Towards using Hopfield Networks for the Identification of Therapeutic Targets for Cancer

Resumen

Cancer currently constitutes both a national and worldwide health problem for the human population. General scientific effort has been directed towards improving targeted and personalized therapy, which are characterized by making an intelligent use of the patient’s genomic blueprint in order to make more informed treatment decisions. NIH’s project, Genomic Data Commons (GDC), provides an openly available online data repository which stores great diversity of cancer related data from different cases (patients). This study leverages patterns found in gene expression data for the identification of potential therapeutic targets. Data was restricted to highly expressed genes from cases of breast invasive carcinoma. A Hopfield network (a type of recurrent neural network) was used for clustering purposes. Preliminary tests subdivided the cases into two clusters, where four highly expressed genes better characterized one cluster from the other. Some of these genes are reported in the literature, indeed, as biomarkers for breast cancer. These results suggest that attractor states of Hopfield networks might provide means for discovering or better understanding potential therapeutic targets when treating a particular cancer subtype.

Index Terms—Cancer, Therapeutic Targets, Neural Networks, Machine Learning, Pattern Recognition, Bioinformatics.