Nanobody CDR Infilling Dataset

Dataset Overview

This dataset helps with completing missing Complementarity Determining Regions (CDRs) in nanobody sequences. CDRs are the highly variable regions in antibodies that directly contact antigens, and they are crucial for antibody specificity and affinity.

The CDR infilling task aims to predict or generate missing CDR sequences based on framework regions (FRs) and other CDR regions. This is valuable for antibody design, optimization, and understanding antibody-antigen recognition.

Data Collection

The dataset is based on nanobodies with known structures and sequences, collected from the Protein Data Bank (PDB), antibody databases, and published literature. The data has been processed to be suitable for sequence completion tasks, including correct labeling and segmentation of CDR regions.

Dataset Structure

The dataset is split into training, validation, and test sets.

File Format

CSV files contain these columns:

• seq_masked: Nanobody sequence with masked (hidden) CDR regions
• mask_indices: Indices showing the positions of masks
• true_cdr: The original CDR sequence that was masked
• cdr_type: Indicates which CDR is masked (CDR1, CDR2, or CDR3)

Uses and Limitations

Uses

• Develop models that can complete CDR sequences
• Design new nanobody sequences
• Understand the relationship between CDR sequence and structure
• Support antibody engineering and optimization

Limitations

• Highly variable CDR3 regions may have multiple valid completions
• Sequence completion doesn't guarantee functionality
• Not considering 3D structure information may limit prediction accuracy

Evaluation Metrics

Model performance is evaluated using:

• Residue Accuracy (Exact Match): Percentage of correctly predicted amino acid residues
• BLOSUM62 Score (BLOSUM Recovery): Uses substitution matrices to evaluate similarity between predicted and actual sequences