Model Interpretability in Genomics
Deep learning (DL) methods have been widely adopted in genomics for extracting biological insights and model predictions because of their superior performance in predictions and classification tasks through their deep neural network (DNN) architecture. Even though the accuracy and efficiency of these model predictions are the primary goals of DL in genomics applications, the decisions made by these DNNs is also important in genomics toward the goal of understanding cellular and molecular mechanisms. In Machine Learning (ML) and DL, "Model interpretability" refers to how easy it is for humans to understand the decisions made by the model. The more interpretable the models are, the easier is it to understand the model's decisions. In contrast, difficulties in model interpretation limit the practical utility of DL models and reduce confidence in their adoption. However, it’s not easy to interpret DL model behavior in a way that...
