AntibodyGPT

As explained in Domain-Specific Small Language Models

A specialized protein generation model in this chapter for generating therapeutic antibody sequences from an antigen context.

AntibodyGPT is a specialized protein-generation model for creating therapeutic antibody sequences from an antigen sequence. It adapts ProGen2, a decoder-only protein language model, to antibody design. section 8.3

The model family is defined by these practical properties:

• Fine-tunes ProGen2 small, medium, and large variants on 5,000 experimentally resolved antibody–antigen crystal structures. section 8.3
• Offers model sizes based on 151-million-, 765-million-, and 2.7-billion-parameter ProGen2 variants through the Hugging Face Hub. section 8.3
• Uses ANARCI to validate whether generated sequences conform to known antibody structural patterns. section 8.5

Generating candidate sequences

Load the model with ProGenForCausalLM, tokenize a target antigen, and generate multiple sequences using sampling controls such as top-p and temperature. GPU acceleration is required for the demonstrated inference workflow. CUDA memory settings and 8-bit quantization can reduce resource demands; quantization reduced the model from 617 MB to 156 MB and improved average latency by 40%. section 8.3 section 8.5

Figure 8.1 The different options available for the ProGen2 and Antibody Generator model families

Compare the available ProGen2 and AntibodyGPT model-family variants before selecting an inference configuration. section 8.3

How Domain-Specific Small Language Models covers it

How deep it goes

Explains model variants, conditioning, inference, validation, memory management, and quantization, but does not establish biological viability.

Examples

Demonstrates antigen-conditioned generation, sequence decoding, ANARCI validation, CUDA allocation settings, and quantized inference.

What it assumes

Assumes an antigen sequence, Python and PyTorch familiarity, GPU access, and use of a custom ProGenForCausalLM repository.

What you can do after

Apply it via ProGenForCausalLM and Hugging Face checkpoints to generate and validate candidate antibody sequences.

How it connects

Connects ProGen2, protein language models, decoder-only architectures, ANARCI, CUDA memory management, and 8-bit quantization.

liveBookmanning.com

Read Domain-Specific Small Language Models on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Domain-Specific Small Language Models. The section links point to the parts of the book it draws on.