Accepted_test

Last year the EvoAug, an open-source PyTorch package, was published by american researchers from the Simons Center for Quantitative Biology [1]. A common biology problem is insufficient data to train a deep neural network (DNN), causing the neural network to overtrain and exhibit poor generalization ability on new data. The authors of the original paper show that training neural networks using the EvoAug framework, that provides a suite of evolution-inspired data augmentations, leads to better generalization performance and improves efficacy with standard post hoc explanation methods, including filter interpretability and attribution analysis, across prominent regulatory genomics prediction tasks for well-established DNNs.

The purpose of this work is to test the effectiveness and test the framework for various tasks of regulatory genomics.