Ab-Ligity: identifying sequence-dissimilar antibodies that bind to the same epitope

Figures & data

Table 1. Number of positive and negative comparisons in the datasets, based on Ab-Ligity’s definition of similar epitopes

Set	Positive	Negative
Full	719	29,557
CDRH3 ≤ 0.8 set	266	29,519

Table 2. Precision and recall on the full and CDRH3 ≤ 0.8 sets, using the selected paratope similarity thresholds for Ab-Ligity (0.1) and InterComp (0.6), based on Ab-Ligity’s definition of similar epitopes

Method	Ab-Ligity		InterComp
Set	Precision	Recall	Precision	Recall
Full	0.95	0.85	0.92	0.77
CDRH3 ≤ 0.8 set	0.95	0.69	0.86	0.59

Table 3. Precision and recall on the full set, using heavy chain and light chain only homology models and the predicted paratopes on the corresponding single domain

Method	Ab-Ligity		InterComp
Domain	Precision	Recall	Precision	Recall
VH	0.88	0.78	0.74	0.82
VL	0.67	0.89	0.17	0.94

Figure 1. Analysis of anti-lysozyme (HEL) antibodies with dissimilar CDRH3 sequences and highly similar epitopes

(A) Structural superposition of three anti-lysozyme and one anti-TNFRSF4 antibodies co-crystallized with their antigens (lysozyme or TNFRSF4) in white. The three anti-lysozyme antibodies were HyHEL-26 (1NDM_BA_C), HyHEL-10 L-Y50F mutant (1J1O_HL_Y) and HyHEL-63 (1NBY_BA_C); and the anti-TNFRSF4 antibody is 3C8 (6OKM_HL_R). The antigens from the three anti-lysozyme antibody crystal structures are aligned. The legend shows the colors of the antibodies with their PDB codes followed by the heavy-light chain and antigen chain identifiers, separated by (‘_’). The CDRH3 sequences are displayed next to the respective antibody identifiers. Crystal paratope residues within the CDRs are in bold, and Parapred-predicted paratopes within the CDRs are underlined. The other five CDR sequences are listed in Supplementary Table S15. (B) Heatmaps of CDRH3 sequence identity and Ab-Ligity paratope similarity. The row and column labels correspond to the structures shown in (A). Ab-Ligity paratope similarity is calculated on the antibody model and predicted paratope as outlined in the Methods section. Pairs of antibodies with CDRH3 identity of >0.80 would have been considered similar by sequence-based metric. For Ab-Ligity, a similarity score of >0.1 suggests that the antibodies bind to highly similar epitopes.

Figure 2. Analysis of two anti-gp120 antibodies. VRC01 antibody (PDB code and chain IDs: 4LSS_HL_G) is colored in blue; VRC07 (PDB code and chain IDs: 4OLU_HL_G) is shown in pink. The gp120 core antigen is displayed as white surfaces and the superimposed antibodies are in cartoons. The CDRH3 loops are in solid shades of the cartoon representation. The PDB code, heavy and light chain ID, and antigen chain ID are separated by (‘_’) and listed in the legend with the corresponding CDRH3 sequences. CDRH3 sequences are shown by aligning their IMGT positions and ‘-’ indicates a gap in the alignment according to the IMGT numbering scheme.²⁹ Crystal paratope residues within the CDRs are in bold, and Parapred-predicted paratopes within the CDRs are underlined. The other five CDR sequences are listed in Supplementary Table S16. The Ab-Ligity paratope similarity score for the pair is listed

Figure 3. The Ab-Ligity workflow

(A) Binding site residues within 4.5 Å of the binding partner are tokenized as stated in Table 4. (B) All distances between tokenized points are calculated and hashed into 1.0 Å -wide distance bins for both the paratopes and epitopes. (C) Each pair of tokens is given a unique hash code. (D) A six-character hash code is generated for each triplet. (E) The hashes of a binding site are stored in a frequency table.

Table 4. Residue groupings for tokenization

Group	Residues
Aliphatic	Glycine (G), Alanine (A), Valine (V), Leucine (L), Isoleucine (I), Proline (P)
Hydroxyl	Serine (S), Threonine (T)
Sulfur	Cysteine (C), Methionine (M)
Aromatic	Phenylalanine (F), Tyrosine (Y), Tryptophan (W)
Acidic	Aspartic acid (D), Glutamic acid (E)
Amine	Asparagine (N), Glutamine (Q)
Basic	Histidine (H), Lysine (K), Arginine (R)

Lefranc MP, Giudicelli V, Duroux P, Jabado-Michaloud J, Folch G, Aouinti S, Carillon E, Duvergey H, Houles A, Paysan-Lafosse T, et al. IMGT®, the international ImMunoGeneTics information system 25 years on. Nucleic Acids Res. 2014;43(D1):D413–D422. doi:10.1093/nar/gku1056.

 PubMed Web of Science ®Google Scholar