GDPa5: Antibody developability dataset of 160 clinical and non-clinical VHHs

Contains developability assay data for 160 VHH-Fc constructs drawn from PLAbDAb-nano, including 36 clinical-stage nanobodies, as described in our preprint Predicting VHH-Fc Developability from Large-Scale IgG Data.

Example usage

Using pandas:

import pandas as pd

# Login using e.g. `hf auth login` to access this dataset
df = pd.read_csv("hf://datasets/ginkgo-datapoints/GDPa5/GDPa5_20261102.csv")

Using Hugging Face datasets:

from datasets import load_dataset

# Login using e.g. `hf auth login` to access this dataset
ds = load_dataset("ginkgo-datapoints/GDPa5")

Data processing

The main table in this data (GDPa5_20261102.csv) has one row per construct. It joins the "Sequences" sheet of the full datasheet with the "Assay Data - average" sheet, which reports the mean, standard deviation, and count of usable replicates for each assay.

• Averages use replicates flagged Usable; AC-SINS additionally requires High measurement confidence.
• HPLC assays (HAC, HIC, SEC, SMAC) are single-injection per antibody, so their _std is blank and _count is 1.
• in_model_cohort marks the 100 constructs meeting the modeling quality threshold (sec_%monomer >= 80). Assay data is reported for all 160 constructs either way.
• Thermostability values come from the 0.5 degC/min nanoDSF re-run, not the original 1 degC/min run.

Antibodies

All constructs are VHHs fused to a human IgG1 Fc (identical Fc sequence across the panel). origin distinguishes clinical-stage therapeutic nanobodies (36) from discovery library nanobodies (124).

Developability assays

1. Thermostability by nanoDSF at a 0.5 degC/min ramp rate (differs from the 1 degC/min run reported in GDPa1)
2. Colloidal stability by SMAC
3. Aggregation by SEC (% monomer)
4. Hydrophobicity by HIC
5. Heparin binding by HAC
6. Self association by AC-SINS at pH 7.4 and pH 6.0
7. Polyreactivity against CHO cell lysate and ovalbumin
8. Polyreactivity against baculovirus particles (BVP) in two blocking buffers (SuperBlock and BSA)

Full Datasheet

Our full datasheet in Excel format (GDPa5_20261102_full.xlsx) contains the following information:

• Definitions of column headers in other datasheets
• Antibody sequences
• Assay data summary statistics with average, standard deviation, and replicates for each assay
• Assay data in "tidy data" format with one row per replicate, including usability, confidence, and QC annotations
• Model summary: train and test metrics for the models in the preprint (IgG-to-VHH transfer and VHH leave-one-cluster-out cross-validation), with dataset provenance and leakage flags. It covers only the 100 constructs in the model cohort.
• Versioning

Changelog

• November 2nd, 2026 (v5.2): Added the model summary.
• August 11th, 2026 (v5.1): Added the per-replicate raw assay sheet; averages recomputed from the raw data; assay data reported for all 160 constructs with an in_model_cohort flag.
• June 18th, 2026 (v5.0): Initial data drop.

Contact

For more information on this data, see our website at https://datapoints.ginkgo.bio/, or contact us at datapoints@ginkgobioworks.com for specific questions about the data.