The primary challenge in modern antibody engineering has shifted from simple target binding to the practicalities of manufacturing. Even if an antibody binds to a pathogen, it often fails due to aggregation or structural instability. HCAbLM addresses this by learning the specific biological grammar of human heavy-chain-only antibodies (HCAbs), a feat conventional general-purpose protein models struggle to replicate due to a lack of specialized training data.
Nona Biosciences Debuts AI Model for Human Antibody Development
Cambridge-based Nona Biosciences has unveiled HCAbLM, the first language model trained exclusively on fully human heavy-chain-only antibodies. By analyzing a massive repertoire of over 31 million sequences, the model aims to solve the industry’s persistent manufacturing hurdles by predicting the stability and developability of potential drug candidates before they reach labs.
In comparative performance tests, the 366-million-parameter model outperformed significantly larger systems, including Meta’s 6-billion-parameter ESM-6B, as well as existing tools like IgLM and AbLang. The model successfully predicted critical drug characteristics such as thermal stability and purity levels measured via chromatography. According to Nona Biosciences CEO Dr. Di Hong, this breakthrough allows the company to translate complex sequence data into reliable predictions, effectively bridging the gap between initial discovery and clinical-grade production.
Comments (0)
No comments yet. Be the first!