Model interpretability methods in genomics
The field of genomics has garnered so much attention lately because of the advances in high-throughput methods, such as next-generation sequencing (NGS), and other omics technologies such as proteomics, metabolomics, and so on. This has resulted in abundant data such that researchers are in a dilemma about how to use this. The DNN methods showed superior performance compared to the state-of-the-art conventional methods in many genomics applications in medical research, especially in imaging tasks, tumor identification, antibody discovery, motif finding, genetic variant detection, and chromatin interaction, to name a few. However, the major complaint from DNN architectures is that they are black-box models. What that means is that we don’t know how these models made decision on a given dataset. To make predictions with a DL model, the input data is passed through several layers of a DNN, each layer containing several nodes that have...
