Application of machine learning to determine the characteristics of adjacent normal tissues in liver cancer

This study applies machine learning methods to gene expression data from normal tissue of patients with liver cancer to predict whether this tissue is 'healthy', 'cirrhotic' (liver damage), 'non tumor', or 'tumor'. The method is based on using Principle Component Analysis (PCA) combined with the Regularized Least Squares (RLS) classifier. The results show a high accuracy with 10-fold cross validation for discrimination among tissue types. Results indicate the capability of gene expression profiling to successfully discriminate between tumor tissue and normal tissue, however there is a clear and strong overlap between non-tumor tissue and cirrhotic tissue. Further, we used the same classification model to predicate the probability of detecting each class separately. Tumor gene expression can be predicated successfully