Recurrent Neural Networks in Genomics

Deep learning (DL) models are so versatile that they can adapt to any input data distribution and, at the same time, generalize very well to previously unseen data. A variety of deep neural network (DNN) architectures have been designed to suit a particular task. For example, we saw how feedforward neural networks (FNNs) are good at making predictions from structured data, such as tabular data, in Chapter 4, Deep Learning for Genomics. We also saw how convolutional neural networks (CNNs) are good at making predictions from unstructured data such as images, audio, text, and DNA sequence data; we saw this in Chapter 5, Introducing Convolutional Neural Networks for Genomics. But what about sequential data? If you look around, we are currently flooded with a lot of sequential data. Some examples include financial data and DNA sequences. The most important type of sequential data is the time series data, which is a series of data points listed in time...