W34: Automated Machine Learning

Machine learning and deep learning has been applied to various problems in genomics and biology, including DNA sequence modeling, gene expression analysis, drug discovery, and protein structure prediction. One of the key aspects of successful application of deep learning in these fields is the proper selection of neural network architecture through automated machine learning (AutoML).

AutoML refers to the process of automatically searching for the best neural network architecture, and more broadly, hyperparameter tuning, model selection and data preprocessing. These techniques are crucial for achieving high performance, while properly leveraging them can significantly improve the generalization of the models and ease the model building process.

In this workshop, we will cover the basics of applying a convolutional neural network (CNN) to model genomic sequences. Building on top of this, we will explore how different network architectures and hyperparameters affect the model’s performance. In the second half of the workshop, we will introduce how this manual tuning of models can be replaced with a Python package. We wrap up the tutorial by interpreting the biological insights we can learn from this CNN model.