7 Generating Protein Structures

 

This chapter covers

  • How to use LLMs to generate protein structures.
  • How to use LLMS to generate crystal structures.
  • How to optimize inference performance of LLMs for chemistry

7.1 Application of Transformers in Chemistry

Chapter 6 was about practical examples of domain specific small language models for Python code generation and how to optimize them to run on commodity hardware (potentially also in a laptop). The task covered in that chapter has been chosen to make easier the showcase of how to overcome inference challenges by applying some of the tips and tricks to improve models’ performance, quality, etc., explained in the first part of this book, as every reader here is a domain expert in Python programming. Before moving to the next chapters of the book, where more advanced techniques about inference optimization and model quantization would be introduced, let’s provide further evidence that the benefits of the techniques explained in the initial chapters aren’t restricted to coding assistance tasks only, by applying them to other specialized small language models in the chemistry domain. If you aren’t familiar with this domain, don’t worry, the chemistry part would be kept at very high level (just the concepts you as a reader would need to know to be in a condition to follow the technical explanation of the proposed models and the applied inference optimization strategies).

7.2 From natural language to protein structures

7.3 Antibody generation with a small language model

7.4 From CIF files to crystal structures

7.5 Summary