TABLE 3.
Comparison of SPACE2, SPACE1, and clonotyping.
| Dataset | CoV-AbDab | Training set | ||||||
|---|---|---|---|---|---|---|---|---|
| Method | SPACE2 | SPACE1 | Clonotyping | SPACE2 | SPACE1 | Clonotyping | ||
| VH | Fv | VH | Fv | |||||
| Fraction of antibodies modeled | 1.0 | 0.71 | - | - | 1.0 | 0.7 | - | - |
| Fraction of consistent clusters | 0.87 | 0.87 | 0.98 | 0.99 | 0.63 | 0.64 | 0.84 | 0.83 |
| Fraction of clustered antibodies in consistent clusters | 0.82 | 0.81 | 0.97 | 0.99 | 0.57 | 0.58 | 0.75 | 0.79 |
| Multiple-occupancy clusters | 1,811 | 1,165 | 1,191 | 970 | 480 | 314 | 361 | 303 |
| Antibodies in multiple-occupancy clusters | 6,271 (59%) | 4,010 (37%) | 4,045 (38%) | 2,754 (26%) | 1,446 (47%) | 935 (31%) | 1,060 (35%) | 793 (26%) |
| Antibodies in consistent multiple-occupancy clusters | 5,126 (48%) | 3,255 (30%) | 3,916 (37%) | 2,733 (25%) | 823 (27%) | 538 (18%) | 797 (26%) | 628 (21%) |
| Fraction of clusters containing >1 VH-clonotypes | 0.81 | 0.71 | 0 | 0 | 0.55 | 0.58 | 0 | 0 |
| Mean CDRH3 sequence identity | 0.54 | 0.57 | 0.88 | 0.88 | 0.66 | 0.67 | 0.86 | 0.87 |
The original SPACE1 algorithm was evaluated at an RMSD threshold of 1.25 Å. Two protocols were used for clonotyping (see Materials and Methods). VH-clonotyping is restricted to genes and sequence of the heavy chain. Fv-clonotyping considers both heavy and light chains. For the two metrics of number of antibodies in multiple-occupancy clusters and number antibodies in epitope-consistent multiple occupancy clusters, a percentage is shown additionally, indicating the percentage of antibodies in the dataset. The most important performance metric to consider when comparing different epitope profiling methods is the number of antibodies in epitope/domain-consistent multiple-occupancy clusters as high accuracy or coverage metrics individually may not indicate good performance. The fraction of epitope/domain-consistent clusters containing more than one VH-clonotype and the mean CDRH3 sequence identity observed within epitope/domain-consistent clusters are also given. The best result for each metric is highlighted in bold.