Use SPACE2 Online

SPACE2: Structure-Based Antibody Clustering for Computational Epitope Profiling

SPACE2 is a structure-based method for grouping antibodies that are likely to bind the same epitope. It was developed by Fabian C. Spoendlin, Brennan Abanades, Matthew I. J. Raybould, Wing Ki Wong, Guy Georges and Charlotte M. Deane at the Oxford Protein Informatics Group, University of Oxford, and Roche Pharma Research and Early Development in Penzberg, Germany. It updates the earlier SPACE1 approach and is described in Frontiers in Molecular Biosciences (2023).

In antibody discovery, deep sequencing of repertoires produces thousands of candidates, and knowing which ones target the same site on an antigen is valuable. The standard shortcut, clonotyping, groups antibodies by sequence features such as shared V genes and CDR3 identity. It is accurate, but it only recovers antibodies that look alike in sequence and therefore misses many that bind the same epitope with very different sequences.

SPACE2 instead predicts the structure of each antibody Fv region and compares the shapes of the binding loops, which lets it link antibodies that sequence-based clonotyping would treat as unrelated.

How SPACE2 Clusters Antibodies by Structure

1. Predict Fv structures: Models of the paired heavy and light chain variable regions are generated with ABodyBuilder2.

2. Group by CDR length: Models are split into groups that share identical lengths for all six CDRs, and comparison happens only within each group.

3. Measure loop similarity: Models are aligned on framework C-alpha atoms, and a pairwise matrix of C-alpha RMSDs over CDR loop residues is computed.

4. Cluster: Agglomerative clustering with complete linkage and a 1.25 A threshold forms the final epitope-level groups.

The authors tested eight clustering algorithms on the training set. Agglomerative clustering, OPTICS-xi and K-means performed best, and agglomerative clustering was chosen because it produced larger epitope-consistent clusters. Using all six CDRs worked best: a heavy-chain-only variant lost accuracy and a paratope-only variant gave fewer consistent clusters.

Results: Broader Epitope Coverage Than Clonotyping

CoV-AbDab comparison

Diversity within clusters

81% of domain-consistent CoV-AbDab clusters contained more than one VH clonotype, and the mean CDRH3 sequence identity inside consistent clusters was 54%, compared with 88% for clonotyping. Twenty-six functionally consistent clusters spanned multiple species (human, mouse and rhesus macaque), compared with 18 for SPACE1, and clonotyping cannot link across species. Held-out tests gave 100% consistent clusters on an anti-lysozyme set of 53 antibodies and 85% on the CoV-AbDab test set.

What is Tamarind Bio?

Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.

The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:

• A user-friendly graphical interface for setting up and launching experiments

• A robust API for integration into existing research pipelines

• An automated system for managing and scaling computational resources

Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.

Accelerating Discovery with SPACE2 on Tamarind Bio

• Epitope binning from sequence data: Group large antibody panels into likely epitope clusters without experimental competition assays.

• Finding diverse binders: Identify antibodies with different sequences that probably share a target site, including across species.

• Repertoire mining: Cluster deep-sequencing hits to prioritize one representative per epitope group for testing.

• Antibody panel triage: Compare candidate sets against known epitope-annotated antibodies to guide selection.

How to Use SPACE2 on Tamarind Bio

1. Open the tool: Log in at tamarind.bio and open SPACE2.

2. Provide antibodies: Supply paired heavy and light chain Fv sequences or models, as the tool exposes. The SPACE2 repository expects IMGT-numbered PDB models with chain H for heavy and L for light chains.

3. Set the cutoff: The paper uses a 1.25 A RMSD threshold with agglomerative clustering; adjust only if the tool exposes it.

4. Run the job: Launch clustering and wait for completion.

5. Download results: Retrieve the table assigning each antibody to a CDR-length group and a structural cluster.

6. Interpret clusters: Treat antibodies in the same cluster as candidates for the same epitope, then validate experimentally.

7. Use downstream: Select representatives per cluster for expression, binding assays or further engineering.

Things to Keep in Mind

• Clonotyping is more accurate when it does cluster (about 98 to 99% consistent on CoV-AbDab, versus 87% for SPACE2), but it covers far fewer antibodies.

• Only antibodies with identical CDR lengths are compared, so convergence across different lengths can be missed.

• Similar CDR shapes can still bind different epitopes if side-chain chemistry differs substantially.

• No clusters mixing antibodies and nanobodies were found, and coverage was low on the Ebola set (15%).

• Structure prediction is the rate-limiting step, and epitope labels for several benchmarks were only domain-level.

Source: Spoendlin FC, Abanades B, Raybould MIJ, Wong WK, Georges G, Deane CM. Improved computational epitope profiling using structural models identifies a broader diversity of antibodies that bind to the same epitope. Frontiers in Molecular Biosciences 2023; 10. DOI: 10.3389/fmolb.2023.1237621. Code: github.com/oxpig/SPACE2.