The Use of the Random Number Generator and Artificial Intelligence Analysis for Dimensionality Reduction of Follicular Lymphoma Transcriptomic Data

Follicular lymphoma (FL) is one of the most frequent subtypes non-Hodgkin lymphomas. This research predicted prognosis 184 untreated follicular patients (LLMPP GSE16131 series), using gene expression data and artificial intelligence (AI) neural networks. A new strategy based on random number generation was used to create 120 different independent multilayer perceptron (MLP) solutions, 22,215 probes were ranked according their averaged normalized importance for predicting overall survival. After dimensionality reduction, final network architecture included (1) newly identified predictor genes related cell adhesion migration, signaling, metabolism (EPB41L4B, MOCOS, SPIN2A, BTD, SRGAP3, CTNS, PRB1, L1CAM, CEP57); (2) international prognostic index (IPI); (3) other relevant immuno-oncology, immune microenvironment, checkpoint markers (CD163, CSF1R, FOXP3, PDCD1, TNFRSF14 (HVEM), IL10). The performance this good, with an area under curve (AUC) 0.89. comparison machine learning techniques (C5 tree, logistic regression, Bayesian network, discriminant analysis, KNN algorithms, LSVM, trees, SVM, tree-AS, XGBoost linear, CHAID, Quest, C&R forest, network) also made. In conclusion, survival a high accuracy.