Antibody Developability · ginkgo-datapoints
GDPa1
A biophysical developability benchmark providing experimental assay measurements for 242 antibodies.
What GDPa1 is
GDPa1 is an antibody developability dataset created by Ginkgo Datapoints. It provides comprehensive biophysical profiling across a panel of 242 antibodies to support machine learning and predictive model training for therapeutic antibody candidate selection. Poor biophysical properties, such as high aggregation propensity or poor colloidal stability, regularly cause antibody drug candidates to fail during downstream development.
The dataset measures key developability attributes across multiple experimental assays, capturing properties including expression titer, purity, aggregation, thermostability, colloidal stability, hydrophobicity, non-specific binding, and self-association. It links these quantitative measurements directly to antibody sequences and includes computational cross-validation splits to facilitate standardised benchmarking.
How it was made
Antibodies were expressed primarily in HEK293F cells and purified using Protein A chromatography prior to developability assessment, with a subset of 20 IgGs produced in ExpiCHO. Candidates measured via dynamic light scattering interaction parameter (DLS-kD) received an additional size-exclusion chromatography (SEC) polishing step. Measurements were captured across 10 distinct assay formats: Valita titer, reduced capillary electrophoresis (rCE-SDS) purity, SEC aggregation, nanoDSF and DSF thermostability, standup-monolayer adsorption chromatography (SMAC) colloidal stability, hydrophobic interaction chromatography (HIC), heparin affinity chromatography (HAC), affinity-capture self-interaction nanoparticle spectroscopy (AC-SINS), bead-based polyreactivity against CHO SMP and ovalbumin, and DLS-kD.
The primary dataset compiles these replicate measurements into an averaged table derived from the first production batch, using the median across all experimental replicates. It provides pre-computed 5-fold cross-validation splits: random folds, hierarchical clustering based on MMseqs2 pairwise sequence identity, and hierarchical clustering stratified by IgG isotype to balance subclass representation across evaluation folds. Supplementary files contain replicate-level tidy data, ABodyBuilder3 predicted 3D structures, and literature comparisons.
Key facts
- Number of antibodies
- 242
- Number of assays
- 10
- Expression hosts
- HEK293F (all), ExpiCHO (20 IgGs)
- Data format
- CSV, Excel
- Modality
- Tabular, text
- Structures included
- ABodyBuilder3 predicted structures
What you can use it for
- Developability Machine Learning Models
- Train and evaluate models that predict biophysical properties such as hydrophobicity, thermal stability, or polyreactivity directly from antibody primary sequences.
- Isotype-Stratified Benchmarking
- Benchmark machine learning algorithms across standardized 5-fold sequence clusters stratified by IgG isotype to prevent data leakage between closely related sequence families.
- Biophysical Assay Correlation
- Investigate relationships between diverse experimental developability metrics, such as comparing AC-SINS self-association against SMAC colloidal stability or nanoDSF against DSF thermostability.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.Strengths and limits
Good at
- Measures a consistent panel of 242 antibodies across 10 distinct physical and biochemical assays.
- Provides pre-calculated sequence-clustered and IgG isotype-stratified 5-fold cross-validation splits.
- Includes both raw replicate-level data and median-averaged values alongside standard deviations.
- Incorporates ABodyBuilder3 structural models alongside sequence information.
Watch out for
- DLS-kD measurements are limited to only 10 antibodies rather than the full panel.
- Different production batches vary in their constant regions, requiring the main averaged table to use only the first batch.
Questions
Can I look inside GDPa1 without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is GDPa1 under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
What cross-validation split is recommended for reporting benchmark results?
The recommended split is hierarchical_cluster_IgG_isotype_stratified_fold, which uses MMseqs2 pairwise sequence identity to cluster related antibodies while maintaining balanced IgG subclass representation across folds.
Which developability assays are included in GDPa1?
The panel includes Valita titer, rCE-SDS purity, SEC aggregation, nanoDSF and DSF thermostability, SMAC colloidal stability, HIC hydrophobicity, HAC heparin binding, AC-SINS self-association, bead-based polyreactivity (CHO SMP and ovalbumin), and DLS-kD.
How were the antibodies expressed and purified?
All antibodies were expressed in HEK293F and purified using Protein A chromatography, with a subset of 20 IgGs also produced in ExpiCHO. Antibodies tested on DLS-kD underwent an additional polishing SEC purification step.
How to cite it
PROPHET-Ab: A high-throughput platform for biophysical antibody developability assessment to enable AI/ML model training. bioRxiv, 2025. https://doi.org/10.1101/2025.05.01.651684v1
Written from GDPa1's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.