We’re excited to share that Eli Lilly and Company (Lilly) has announced the release of their TuneLab VHH developability model called NanoLab built on robust data generated by BigHat's Milliner™ platform. TuneLab is a collaborative AI platform for drug discovery, giving participating companies access to models trained on decades of Lilly research results and letting them contribute data to improve those models via federated learning. BigHat is proud to have generated data behind TuneLab’s VHH developability model.
Lilly TuneLab is part of Lilly Catalyze360, alongside Lilly Ventures, Lilly Gateway Labs, and Lilly ExploR&D, which together support biotech innovation by providing access to strategic capital, lab space and technology, and research and development capabilities.
Antibody developability models are key to advancing next-generation therapeutics: they capture the physicochemical and biophysical properties that determine whether an antibody can be manufactured, stored, and administered as a therapeutic. The lack of high-volume, high-quality developability data is a major bottleneck to realizing AI’s potential in drug discovery. Since founding in 2019, BigHat has built Milliner™ as an end-to-end platform: a high-speed wet lab powered by proprietary cell-free protein synthesis (CFPS) and automation, a custom LIMS++ piping lab data into a structured data lake in real time, and a proprietary trained model zoo built from this data.
Milliner™ is purpose-built for AI-driven workflows. Rigorous reagent manufacturing, ML-guided reaction conditions, and precise DNA template design let us express 90% of BigHat’s AI-generated designs, and for sequences that push the limits of any expression system, we build targeted ML models on the fly. We express, purify, and characterize thousands of antibodies weekly, with deep functional and developability characterization on a one-week turnaround. This speed and throughput allows us to tune our models in real time and make continuous progress on our therapeutic programs.
Speed matters, but data quality matters more. Having characterized hundreds of thousands of antibodies, our team has deep experience spotting inconsistencies before they reach a training set. The data pipelines we’ve built automatically evaluate each metric and, for trace-based assays like CE-SDS, nanoDSF, and SEC, assess peak shape and inflection points that support consistent, well-informed calls.
For the TuneLab dataset, BigHat characterized over 2500 VHHs across 6 one-week design rounds, hitting around 90% field-level completeness across yield, monomeric purity (HPLC-SEC), CE-SDS purity, and thermal properties, delivered on time. ML models are only as good as the data they’re trained on, and data integrity is something we take seriously. A sample that didn’t express well enough for downstream assays or had an unclear thermostability curve was flagged rather than discarded. Accurately representing even the negative data can meaningfully improve the power of our models. On replication, nearly all measured metrics land at CVs between 2-5%.
The TuneLab team has been a great partner on this project, and we look forward to seeing how the Lilly TuneLab models advance drug discovery. As a newly onboarded TuneLab member ourselves, BigHat is looking forward to increasingly visible industry demonstrations of the impact of data on patient outcomes. Thanks to the Lilly and BigHat teams who made this work possible!
