A variable region is the sequence-diverse portion of an antibody that determines much of its antigen-binding specificity. It physically docks onto an antigen, such as a viral spike protein. section 12.3.4
For computational screening, analyze the antibody heavy-chain sequence containing the variable region and classify its likely viral target. Sequence patterns can distinguish antibodies targeting HIV-1 from those targeting SARS-CoV-2. section 12.3.4
Listing 12.3 Transformer-based Classifier for Antibody Sequences
class AntibodyClassifier(nn.Module):
def __init__(
self, vocab_size, padding_idx, embedding_dim,
num_layers, num_heads, ffn_dim, dropout, num_classes,
):
super().__init__()
self.embedding = nn.Embedding( #A
vocab_size, embedding_dim, padding_idx
)
self.pe = PositionalEncoding(embedding_dim)
self.encoder = Encoder(
num_layers=num_layers,
embedding_dim=embedding_dim,
num_heads=num_heads,
ffn_dim=ffn_dim,
dropout=dropout,
)
self.classifier = nn.Sequential( #B
nn.Linear(embedding_dim, embedding_dim),
nn.LayerNorm(embedding_dim),
nn.LeakyReLU(),
nn.Linear(embedding_dim, num_classes),
)
def forward(self, batch):
input_ids = batch["input_ids"]
attention_mask = batch["attention_mask"]
x = self.embedding(input_ids)
x = self.pe(x) #C
x = self.encoder(x, attention_mask)
x = self.mean_pooling(x, attention_mask) #D
logits = self.classifier(x) #E
return logitsA Transformer classifier converts antibody sequences into representations and predicts their target class, providing a practical sequence-level use of variable-region information. section 12.3.4
