A B-cell receptor (BCR) sequence dataset collects antibody sequences from immune repertoires with known biological targets. Each BCR is an antibody attached to a B-cell. A practical dataset can label sequences from HIV-1 and SARS-CoV-2 infections for binary antibody classification. section 12.3.4
A representative dataset and screening workflow include:
- 364 sequences total: 182 HIV-1 and 182 SARS-CoV-2 sequences, producing balanced classes. section 12.3.4
- Sequence lengths average 220 amino acids and range from 114 to 480, creating a meaningful length-variation check. section 12.3.4
- Token embeddings, positional encodings, and mean pooling convert each sequence into a vector for classification. section 12.3.4
- Donor source, study protocol, clonotype overlap, gene-family usage, length, and preprocessing can confound apparent performance. section 12.3.4
Antibody screening aid
A transformer classifier ranks whether an antibody sequence resembles an HIV-1 or SARS-CoV-2 binder, helping prioritize candidates for follow-up rather than replacing experimental validation. section 12.3.4
