SUPR-DSF Case Study: AI-Powered Antibody Developability Assessment

Customer

Ginkgo Bioworks, Inc. – A synthetic biology company enabling high-throughput antibody screening and machine learning model development for biopharma R&D.

Application Area

Therapeutic antibody development: Predicting and improving antibody developability properties for better manufacturability, stability, and clinical success rates.

Challenge

Machine learning (ML) models are transforming antibody drug design, but their effectiveness in predicting developability is limited by the size and quality of training datasets. Experimental platforms to generate large-scale, standardised biophysical data are urgently needed.

Solution

Ginkgo Bioworks developed PROPHET-Ab, a high-throughput developability platform powered by automated analytics and advanced biophysical assays. A key component of this platform is nanoDSF, enabled by SUPR-DSF, used to determine antibody thermostability with label-free precision.

Using SUPR-DSF, Ginkgo measured:

• T_m1 and T_m2 transitions (CH2 and Fab/CH3 domains) across 246 antibodies
• Onset of unfolding (Tonset) to distinguish subtle stability differences
• Compared nanoDSF vs dye-based DSF, confirming nanoDSF's superior accuracy and non-interfering nature
  
  

Results

Thermal Stability Profiles Across IgG Subclasses and Comparison of nanoDSF vs DSF

SUPR-DSF revealed that IgG4s generally exhibit a lower median Tonset and Tm1, suggesting reduced overall thermal stability compared to IgG1 and IgG2. In addition, nanoDSF consistently reported higher Tm values than dye-based DSF methods, confirming its superior sensitivity and accuracy.

Antibody Titer and Monomeric Purity by IgG Subclass

The SEC and rCE-SDS data indicate that IgG1s and IgG2s typically display higher monomer content than IgG4s. This informs early developability decisions and subclass optimisation strategies.

AC-SINS as a Surrogate for DLS-kD in Self-Association Assessment

AC-SINS measurements correlate strongly with DLS-kD results, supporting its use as a reliable and high-throughput method for early assessment of antibody self-association tendencies.

Performance of ML Models Trained on Developability Data

As dataset size increases, model performance improves, supporting the use of SUPR-DSF-derived data for predictive ML workflows. ESM-2 embedding models demonstrate higher Spearman correlations with assay outcomes compared to one-hot encoding.

Impact

✔ Label-Free Methodology: SUPR-DSF uses intrinsic fluorescence for high sensitivity and accuracy without interference from external dyes.

✔ Automation-Friendly: seamlessly integrates into drug discovery workflows, facilitating rapid developability screening.

✔ ML-Ready Data: Supports advanced predictive modeling for better candidate selection and risk mitigation.

Citation

Data and figures adapted from Arsiwala et al., 2025. “A high-throughput platform for biophysical antibody developability assessment to enable AI/ML model training.” bioRxiv. 

Read the full paper

Learn more