Antibody generation

A fine-tuned protein large language model addresses inefficiencies in antibody discovery by generating diverse and functional human antibodies against various targets, overcoming limitations of conventional methods and AI-based techniques that rely on structure-based training data.

WO2025245380A1 Publication Date: 2025-11-27VANDERBILT UNIV
1 Cites 0 Cited by

Patent Information

Application Number
PCT/US2025/030650
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-05-22
Filing Date
2025-05-22
Publication Date
2025-11-27

AI Technical Summary

Technical Problem

Conventional antibody discovery methods are inefficient, costly, and limited by the need for large amounts of practical trial and error, while existing AI-based techniques struggle with complex biological datasets and require antibody-antigen structures that are less available than sequences, limiting their training data.

Method used

A computer-implemented method using a fine-tuned protein large language model (PLM) for autoregressive training on antigen-antibody sequences, enabling the generation of human antibodies without requiring antibody-antigen structures, and incorporating filtering and ranking criteria for humanness, germline identity, and mutational load.

Benefits of technology

The method generates diverse and functional human antibodies with demonstrated binding specificity against targets like SARS-CoV-2, H5N1, and RSV-A, showcasing efficient and accurate antibody design capabilities.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

An example method of training a large language model for antibody generation includes: receiving a training corpus comprising a plurality of antigen amino acid sequences and a corresponding plurality of antibody sequences, generating a plurality of training strings, where each training string comprises an antigen sequence and a corresponding antibody sequence, fine-tuning a pretrained protein language model by autoregressively training on the plurality of training strings, and outputting a fine-tuned protein language model configured to output human antibodies in response to receiving an input string comprising an antigen.
Need to check novelty before this filing date? Find Prior Art