SAbBox in 2023: ImmuneBuilder and more! | Oxford Protein Informatics Group

For several years now, we have distributed the SAbDab database and SAbPred tools as a virtual machine, SAbBox, via Oxford University Innovation. This virtual machine allows a user to utilise the tools and database locally, allowing for high-throughput analysis and keeping confidential data within a local network. Initially distributed under a commercial licence, the platform proved popular and, in 2020, we introduced a free academic licence to enable our academic colleagues to use our tools and database locally.

Following requests from users, in 2021 we released a new version of the platform packaged as a Singularity container. This included all of the features of SAbBox, allowing Linux users to take advantage of the near bare-metal performance of Singularity when running SAbPred tools. Over the past year, we have made lots of improvements to both SAbBox platforms, and have more work planned for the coming year. I’ll briefly outline these developments below.

ImmuneBuilder

We recently released the ImmuneBuilder apps (ABodyBuilder2, TCRBuilder2, and NanoBodyBuilder2) on our website. Following some development work, these are now also included in both version of SAbBox. The release of these tools has some implications for the rest of the platform. Unlike ABodyBuilder and ABodyBuilder-ML, the ImmuneBuilder tools are end-to-end deep learning methods that predict a structure directly from an amino acid sequence. This greatly simplifies the installation process and allowed us to retire older tools such as the venerable loop modeller FREAD. We have now updated both SAbBox platforms to include both the command line and web application verisons of all three ImmuneBuilder applications, allowing users to simply download their format of choice and run the apps locally.

Hu-mAb Improvements

We recently updated Hu-mAb, our antibody human-ness predition tool, to automatically distinguish between kappa and lambda light chains, and this feature is also included in the newest version of SAbBox. Two wonderful students in the group, Ashley Wong and Alissa Hummer, are currently developing improved classification models for Hu-mAb, which will make the tool faster and reduce its memory and storage footprint, both of which should make high-throughput analysis using SAbBox more efficient in the future.

Local SAbDab Updates

One of the advantages of the SAbBox virtual machine over the container was the ability to trivially allocate a storage volume to the virtual machine, which allowed it to keep a local copy of the SAbDab database up to date. Containers, in contrast, are read-only, necessitating an alternative solution. This year we reconfigured the SAbBox container to look in mounted directories on the host file system for data associated with tools such as SAbBox and Hu-mAb. The advantage of this was twofold: firstly, as the mounted volume is writable, it enabled the SAbDab API inside the container to update and maintain a local copy of the SAbDab data stored on the host, which can then be managed independently of the container file itself. This allows the container to be updated without necessitating any modifications to the local SAbDab data, resulting in a much more lightweight container. Secondly, it enables us to distribute updated data and model files independently of the container, such as updated Hu-mAb models. While there is now a little extra work associated with mounting directories when using the container, the result is a much more flexible distribution system which will greatly streamline future updates.

Obtaining SAbBox

All of our tools are freely available as web applications on our website. If you’re interested in trying any of these tools out locally, you can obtain licences for SAbBox through the Oxford University Innovation software store, links to which are included below. Licences are free to academic users, so please do give it a go. If you have any questions or feedback, please don’t hesitate to email us at opig@stats.ox.ac.uk.