liveBook · Manning

Antibody

As explained in Build AI Drug Discovery Pipelines

An antibody is an immune-system protein that specifically recognizes and can help neutralize a pathogen or other antigen.

An antibody is an immune-system protein whose variable heavy-chain region helps determine which antigen it binds. Sequence analysis can support screening antibodies for virus-specific follow-up. section 12.3.4

A Transformer-based antibody classifier follows this workflow:

  • Input antibody heavy-chain sequences labeled HIV-1 or SARS-CoV-2. section 12.3.4
  • Represent residues with token embeddings and positional encodings, then mean-pool them into one sequence vector. section 12.3.4
  • Train a binary classifier with cross-entropy loss and evaluate accuracy, precision, recall, F1, and AUC. section 12.3.4

Virus-target screening

A test run achieved 0.7609 accuracy, 0.8053 AUC, 0.7738 precision, 0.7609 recall, and 0.7580 F1. Predictions below 0.6 confidence can be flagged for review.

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Descends from biological definition through Transformer representation, training, metrics, confidence filtering, and length-stratified evaluation.
Examples
Demonstrates HIV-1 and SARS-CoV-2 sequences, confidence scores, confusion matrices, and short-versus-long sequence accuracy checks.
What it assumes
Assumes sequence labels and basic neural-network concepts while defining antibody regions, supervised classification, and sequence representations.
What you can do after
Diagnose classifier behavior with standard metrics, confidence thresholds, confusion matrices, and sequence-length checks.
How it connects
Connects antibodies with antigens, heavy chains, Transformer models, masked language modeling, protein language models, and supervised classification.

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.