On 5th April 2024, over 60 researchers braved the train strikes and gusty weather to gather at Lady Margaret Hall in Oxford and engage in a day full of scientific talks, posters and discussions on the topic of adaptive immune receptor (AIR) analysis!
The day started with a comprehensive opening keynote from Professor Charlotte Deane [University of Oxford], in which she surveyed her group’s suite of tools for AIR analysis before focusing on recent work on affinity prediction (Chinery et al. 2024), antibody language modelling (AbLang2), inverse folding (AntiFold), structural modelling (ABodyBuilder2/NanoBodyBuilder2/TCRBuilder2), and machine learning extrapolation vs. interpolation (Greenshields-Watson et al. 2024).
After a short break for refreshments, we enjoyed a set of three talks from Dr. Joseph Watson [University of Washington], Dr. Monica Fernández-Quintero [Scripps Institute], and Isabel Elliott [University of Southampton]. Respectively, they covered the history of RFDiffusion and recent efforts to tune it towards AIR design, the role of enhanced sampling molecular dynamics in understanding AIR specificity and biophysics, and how structural biology and Small Angle X-ray Scattering can be combined to design antibodies with bolstered immunostimulatory effector functions.
We then broke for lunch and a poster session in which 18 attendees presented an impressive diversity of work spanning antibody, nanobody, T-cell receptor, and AIR co-receptor characterisations. Discussions were free flowing, and many new connections were made!
The afternoon talks began with a shorter session on T-cell receptor analysis. The first, by Dr. Paula Dobinić [Immunocore], highlighted the use of existing computational approaches to home in immune repertoire- or phage library-derived clones worthy of further investment as well as ongoing efforts to leverage the increased quantity of function-labelled paired sequence data to build new predictive models. James Henderson [University College London] then introduced us to a novel statistical framework for modelling the antigen specificity information content (over baseline) provided by alpha-only, beta-only, and paired alpha-beta sequences.
The final session focused on the multi-parameter nature of biologics drug discovery. Dr. Habib Bashour [University of Oslo] led us through his research outlining the similarities and differences in the computational developability profiles of genetically human antibody therapies and natural B cell receptor repertoires. Dr. Pietro Sormanni [University of Cambridge] then shared a history of his group’s work on antibody/nanobody binder design with enhanced solubility, from grafting approaches to AI-guided molecular engineering via latent representations of human antibodies.
Finally, Professor Peter Tessier [Univeristy of Michigan] wrapped up the conference with a tour de force on joint polyspecificity and self-interaction prediction, from bespoke in vitro assay design through to the development of simple and powerful generalisable models able to identify problematic clones and guide downstream molecular optimisation. Along the way we heard how more accurate structure prediction can help boost model performance and how regions outside of the CDR3 can often drive developability issues, leading to (for example) strong links between germline family and problematic biophysics.
Overall, the day’s talks highlighted the importance of:
- Multi-property optimisation: a wide range of properties are important for biotherapeutic development; we need to move towards optimising multiple simultaneously to improve efficiency and success rates
- Class-specific tools: for certain tasks, models trained or fine-tuned on, for example, antibody data outperform equivalent tools built for general proteins, underscoring the unique structural and sequence features of AIRs
- Considering dynamics: antibodies are not static, and we may not achieve an accurate representation of their properties, including ones related to developability, when looking at only a static structure
- Information density and orthogonality: not all pieces of information have the same level of information value (if understood, this can be leveraged for prediction tasks and the design of ML datasets); sequence and structure capture orthogonal information, for example, related to developability properties
- Baselines: we need robust baselines to understand, and quantify, the true value of new methods
On a personal level, we took away not only a host of scientific learnings, but also useful insights into the effort and steps it takes to run even a one-day seminar! A big highlight of organising a conference is the opportunity to invite people you think are doing exciting work to come and speak. This involved thinking about the direction the field is moving in and trying to capture the key people, scientific advances and research areas – one of the most enjoyable aspects. We also tried as much as possible to invite a diverse group of speakers. A great piece of advice we received was to curate separate lists for different groups (ranging from scientific discipline to career stage to gender, etc.) at a very early point in the process. However, we realise there is always more that could be done to ensure a broad diversity of speaker backgrounds and ideas and would greatly welcome further suggestions that could be implemented at future events.
Moreover, organising AIR-2024 highlighted to us the importance of the venue in an event’s success. One of the most important factors is the layout and its suitability (up there with or even exceeding the location and how nice it is!). LMH not only has a beautiful lecture theatre but also a custom-made versatile adjoining space for breaks, that naturally encouraged people to talk to one another and enhanced the day for many attendees.
In closing, we would like to thank everyone who took part for supporting this inaugural conference and extend special thanks to LMH for their hospitality and AV assistance, and the UK Molecular Graphics and Modelling Society for supporting our concept and offering financial support. We hope to build on the momentum of this meeting with another seminar on the topic in a few years’ time.
Alissa & Matt
