Those familiar with the immunoinformatics work of our group will know of our Observed Antibody Space (OAS) and Observed T-cell Receptor Space (OTS) databases, which compile and consistently process V(D)J transcriptomics datasets from across a range of studies on natural immune repertoires. These have provided a central community resource of training data for adaptive immune receptor (AIR) large language models, as well as references for studies aiming to identify clones related to particular disease states.
We are now delighted to release the final installment of our natural AIR database series: introducing OAS-nano. This project was enabled by key collaborative contributions from Klara Kropivsek Brumat and Ario de Marco at the University of Nova Gorica and our industrial partners at AstraZeneca: Olga Obrezanova, Erwin de Genst, and Rebecca Croasdale-Wood.
Building a robust and useful natural VHH sequence database required several adaptations to the original OAS processing pipeline. For example, VHH immunoglobulin reference gene datasets are much sparser than their antibody equivalents. To maximize sequence recovery, we combined alpaca and Dromedary camel V-gene reference sets. This means entries in OAS-nano that we are confident are bona fide heavy chains variable domains, but that can’t be picked up by species-matched reference immunoglobulin gene due to data sparsity, are still retrieved, providing maximal data for language model applications. Where reads were assigned a V gene call from both species, only the assignment from the species most closely related to the source organism was retained (and this consistency is flagged via a new column in case it matters for your study). Even for species-matched germline gene sets, the appropriate germline gene may be missing from the reference set, so the closest available gene may be assigned instead. To support germline-dependent analyses, OAS-nano keeps the top three germline gene assignments for each sequence, so you can see whether they consistently point to the same gene family, or whether one gene is the clear best assignment.
Another key consideration was that VHH studies use more variable read amplification strategies than antibody sequencing campaigns. The classical approach is to perform nested-primer strategy, whereby both conventional antibody VH and VHH sequences are first amplified using an IGHV3-specific leader primer. Mass separation is then used to enrich for VHH’s followed by a second round of amplication of the framework 1-framework 4 region. Hence two potential confounding factors are introduced: only IGHV3-derived sequences will be amplified, as well as a mixture of bona fide VHH and contaminating VH sequences due to mass separation not always providing two distinct populations. Other datasets in OAS-nano use a hinge-primer protocol, which in the first step selectively amplifies only VHH sequences via their characteristic hinge region sequence. This step is often coupled with a more diverse set of leader primers covering the IGHV1, IGHV3, and IGHV4 V-gene families. The net result would be a distributional shift from the nested-primer protocol; hinge-primer studies by design should not contain any contaminating antibody and will sample a broader diversity of VHH gene origins. In OAS-nano, the amplification protocol is clearly annotated for each dataset, making it easy to identify and account for these experimental biases when building robust model training/eval datasets.
We have also deduplicated each dataset for truncated (trivial) read variants in addition to our standard OAS read QC pipeline. For other ways we have made the datasets in OAS-nano more ML-ready, please see our paper.
Overall, after the read deduplication step, OAS-nano contains 73 million sequences, and includes all publications to date that have released raw, full-length nanobody NGS data to the public domain. This represents 5x the largest previous resource when applying the same redundancy filter.
OAS-nano is the only VHH AIRR database with a searchable web application, which can be found at https://oas-nano.opig.stats.ox.ac.uk. The publication is available in the 2027 Database Edition of Nucleic Acids Research, accessible at https://doi.org/10.1093/nar/gkag1009.
