Summary

In this chapter, we started with understanding the concepts of genomics and how you can store and manage large genomics data on AWS. We also discussed the end-to-end architecture design for transferring, storing, analyzing, and applying ML to genomics data using AWS services. We then focused on how you can deploy large state-of-the-art models for genomics, such as DNABERT, for promoter recognition tasks using Amazon SageMaker with a few lines of code and how you can test your endpoint using code and the SageMaker Studio UI.

We then moved on to understanding proteomics, which is the study of protein sequences, structure, and their functions. We walked through an example of predicting protein secondary structure for protein sequences using a Hugging Face pretrained model with 11 billion parameters. Since it is a large model with memory requirements greater than 220 GB, we explored various memory-saving techniques, such as activation checkpointing, activation offloading, optimizer...