Alchemab announces the publication of AntiBERTa, an antibody-specific machine learning model with multiple applications

BOSTON & CAMBRIDGE, England–(BUSINESS WIRE)–Alchemab Therapeutics, a biotechnology company focused on the discovery and development of natural protective antibodies and immune repertoire-based patient stratification tools, today announced the publication of research demonstrating the potential AntiBERTa (Antibody-specific Bi-directional Encoder Representation and Transformers), a transformer neural network that reads the components of an antibody amino acid sequence, to deeply understand the structure and function of antibody sequences antibody. The article, titled “Deciphering the language of antibodies using self-supervised learning” was published online in the journal Grounds. AntiBERTa is a 12-layer transformer model that provides a contextualized digital representation of antibody sequences and learns biologically relevant information.

“AntiBERTa forms the foundation of Alchemab’s machine learning platform, providing a pre-trained model that is primed for multiple downstream tasks relevant to antibody drug discovery,” said Jake Galson, Ph.D. ., Head of Technology at Alchemab. “We have already demonstrated the utility of AntiBERTa for the prediction of binding sites, which helps us better identify convergent antibodies associated with disease resilience. We are excited to advance our research and leverage our expertise to develop innovative methods of treating diseases in the field of immunotherapy.

The study found that the sequence representations of the B cell receptor (BCR) segregate based on mutational load and the underlying BCR V gene segments used. Importantly, there is a distinct partitioning of naïve B cell-derived BCRs compared to memory B cells, suggesting that functionally important information is captured by the model. Finally, the model recognized pairs of positions in the BCR sequence that form contacts in three-dimensional space. These data demonstrate that AntiBERTa learns various characteristics of BCRs, such as B cell origin, level of activation, immunogenicity and structure.

Dr. Jane Osbourn, PhD, co-founder and chief scientific officer of Alchemab, said: “Our AntiBERTa technology has the potential to transform our ability to understand antibody structure and function and will inform our understanding of antibody paratopes, or sequences of amino acids. including the site at which antibodies bind to antigens. It will also allow Alchemab to continue to develop its unbiased platform to identify new oncology and neurodegenerative targets. Alchemab’s novel approach learns from nature and naturally optimized antibodies and works backwards to uncover the most important targets and pathways involved in modulating disease. This approach has been very successful and has led to the identification of several new drug targets in oncology and neurodegenerative diseases.

About Alchemab

Alchemab has developed a highly differentiated platform that enables the identification of novel drug targets, therapies and patient stratification tools through the analysis of patient antibody repertoires. The platform uses well-defined patient samples, deep B-cell sequencing, and computational analysis to identify convergent protective antibody responses in individuals susceptible but resilient to specific diseases.

Alchemab is building a broad pipeline of protective therapies for hard-to-treat diseases, with an initial focus on neurodegenerative diseases and oncology. The highly specialized patient samples that power Alchemab’s platform are made available through valuable partnerships and collaborations with patient representative groups, biobanks, industry partners and academic institutions.

For more information, visit

Sherry J. Basler