Interpretable model to identify breast cancer information flow based on sparse neural network with tags

Abstract

Along with progressions in the machine learning field, the complexity of model also has increased.This complexity causes the “Black box problem”

which is a mechanistic blackout statement.One trial to solve this black box problem is making the Interpretable model that represents results with reasons why the model makes certain decisions. Here, I developed an interpretable deep learning model, Sparse Neural Network with Tags(SNNT), which mimics the information flow of real world phenomena. According to breast cancer subtyping task in this research, SNNT outperformed other machine learning techniques. In addition, important genes were identified including three new candidate breast cancer subtype risk factors