Revisited the project of elderly EBV lymphomas of CHL or DLBCL done a few weeks ago, and added in a different variation of spread to see since no baseline comparison for fold change top genes if the spread from max-median for stimulated genes to median-minimum in each class of lymphoma could be used as better predictors in a 3 class model using random forest classifier. The results were just as poor with the best class predicted at 75% accuracy for mDLBCL in the training model and about 72% the best score for the CHL class using a subset of the top genes. The duplicates of top genes from extracting the Gene ID from the SPOT.1 feature threw off the model classifier in using all features to predict the model and only subsets could be used of the top genes or batches to predict the class as CHL, pDLBCL, or mDLBCL. The pDLBCL was at best 50% accurate in a subset of genes. Overall best accuracy was 50% prediction using a subset of these top genes.

Atividade CPAD PPGIA UFRPE

Este es mi primer cuaderno :)