This year we (Fabian and Henriette) were invited to speak at the Biologic Summit. The conference took place January 20-22 in San Diego. Fabian presented his work on conformational ensembles of antibodies [1] in the “Data Strategies and the Future of AI Models” track. Henriette presented her work on LICHEN [2], a tool to generate an antibody light sequence for a specific heavy sequence, in the “ML/AI for Biologics Developability, Optimization and de novo Design” track. Below we give some general highlights of the conference, and some talks we enjoyed. We would like to thank the organisers for the opportunity to discuss our research and hear about the latest developments in harnessing ML for design and optimisation of antibodies.  

General feedback

• Industry focused conference. Biologic Summit is strongly attended by industry, making this conference an excellent opportunity to promote your tools/databases, and to connect with companies. The conference is attended by both start-ups as big pharma companies. 

• Medium size conference. With approximately 250 attendees, the Biologic Summit provides a good opportunity to connect with researchers from a wide range of disciplines. Held concurrently with Protein Science and Production Week (PepTalk) and sharing the same venue, the event further benefited from a diverse mix of scientific backgrounds and expertise. 

• Panel and table discussions. Throughout the three days there where various table discussions and panel discussions organised. These are good places to learn about general interest and challenges in the field. 

• Well-organised conference. The conference is well-organised with a clear schedule and enough breaks to recharge and connect. Most talks are scheduled for 30 minutes with around 4 talks per block.

General highlights

• Multispecific antibodies. Bi-specific and multispecific antibodies are of high interest. Formats and constructs designed for binding multiple antigens at the same time were presented as the Antibody Zoo.  

• Immunogenicity and anti-drug antibodies (ADA). Concerns about risks caused by immunogenicity and ADAs were discussed. The Immunogenicity Database Collaborative by Agnihortri and colleagues was presented capturing clinical immunogenicity data of therapeutics [3]. 

• Multi-parameter and multi-modal optimisation. Optimising multiple developability features is crucial for designing a successful therapeutic. Different ways of multi-parameter and multi-modal optimisation were presented.  

• Data storage and legacy data. Standardization of data is important for training machine learning models. Discussions highlighted the best practices for data storage and how historical data can be used, and whether it is worth restoring the data. 

High-throughput data generation and active learning for developing multispecific antibodies

One interesting talk on the design of multispecific antibodies was presented by Winston Haynes from LabGenius. Using a closed-loop platform that integrates machine learning with high-throughput functional screening, they engineered multispecific  antibodies exhibiting highly selective tumour cell killing [4]. A central challenge in developing T-cell engagers for solid tumours is on-target, off-tumour toxicity. Because many tumour-associated antigens are also expressed at lower levels on healthy tissues, the therapeutic window is often narrow, requiring a careful balance between effective tumour targeting and avoidance of healthy cells. To address this limitation, they employed avidity-driven selectivity, leveraging multivalent binding to discriminate based on antigen density. Using HER2 as a case study, they demonstrate how an active learning approach including computational construct selection and high-throughput experimental expression and developability assessment in short repeatable cycles can serve as a general strategy for engineering multispecific biologics. 

Monocolonal antibody discovery directly from serum with cryo-EM

James Ferguson from Scripps presented a talk on capturing target specific antibodies directly from blood serum using cryo-electron microscopy (EM) [5]. Classically serum antibodies have to be sequenced followed by some sort of epitope mapping which takes a lot of time. The methods discussed here skips a lot of these steps which leads to significant speed up. In principle, the workflow has two steps, (1) EM polyclonal epitope mapping and (2) EM “sequencing”, and allows to identify sequences of eptiope specific antibodies within a couple of weeks.

For the epitope mapping step, blood serum from an infected individual is extracted and antibodies isolated which are then incubated with the target antigen. This mixture is then directly imaged in the cryo-EM to obtain low resolution complext structures which depict the binding site and orientation of the antibody. The EM “sequencing” step requires the collection of a high resolution EM structure. If side chains are resolved well a sequence can be fit to the electron maps, however, sometimes computational tools like inverse folding are required to fill gaps.

References

[1] Spoendlin, F. C., Fernández-Quintero, M. L., Raghavan, S. S., Turner, H. L., Gharpure, A., Loeffler, J. R., … & Deane, C. M. (2025). Predicting the conformational flexibility of antibody and T cell receptor complementarity-determining regions. Nature Machine Intelligence, 1-13.

[2]  Capel, H. L., Ellmen, I., Murray, C. J., Mignone, G., Black, M., Clarke, B., … & Deane, C. M. (2026). LICHEN enables light-chain immunoglobulin sequence generation conditioned on the heavy chain and experimental needs. Communications Biology. 

[3] Agnihotri, S., Gonzalez-Nolasco, B., Monian, B., Pattijn, S., Ackaert, C., Wu, P., … & Leventhal, D. S. (2025). The Immunogenicity Database Collaborative (IDC): A Standardized, Publicly Available Database for Clinical Immunogenicity Observations and Insights. bioRxiv, 2025-12. 

[4]  Grace, J., Colin, P. Y., Foxler, D., Haynes, W., Howsham, C., Kassimatis, L., … & van Heeke, G. (2025, December). Engineering multispecific antibodies with complete killing selectivity through the closed-loop integration of machine learning and high-throughput experimentation. Mabs (Vol. 17, No. 1, p. 2598093). Taylor & Francis.

[5] Ferguson, J. A., Raghavan, S. S. R., Peña Alzua, G., et al (2025). Functional and Epitope Specific Monoclonal Antibody Discovery Directly from Immune Sera Using Cryo-EM. Science Advances (Vol. 11, No. 33).