CDD Vault now provides access to Lilly TuneLab, bringing predictive models into your discovery workflows. You can generate predictions for individual molecules or antibodies, evaluate groups of entities, and profile new small-molecule hypotheses directly in the structure editor. Federated learning lets you improve the models with your own data over time, without ever sharing your structures.
What is Lilly TuneLab?
Lilly TuneLab is a first-of-its-kind, collaborative AI-powered drug-discovery and development platform that gives biotech researchers access to powerful AI models trained on decades of Lilly proprietary research data and member data contributions.
As of October 2026, TuneLab models are trained on more than two million preclinical data points from Lilly’s research and more than 675,000 partner-contributed data points. Models continue to improve as TuneLab members contribute new data through federated learning.
Through the CDD Vault integration, you can apply TuneLab’s ChemLab and AbLab models to compounds and antibodies within the context of your existing research data. The small molecule regression and classification models cover more than 30 physicochemical and ADMET endpoints such as solubility, permeability, and hERG binding. The antibody regression models cover more than 17 endpoints and predict critical developability properties like thermal stability, aggregation, and non-specific binding.
A word about security
Through federated learning, TuneLab members help improve shared predictive models using their own experimental data, without sharing raw data with Lilly or with other TuneLab members.
With this release, researchers can participate in TuneLab directly from CDD Vault. CDD hosts the TuneLab federated learning node within CDD Vault’s secure, audited infrastructure, so your compound structures, antibody sequences, and assay results never leave CDD Vault. Models train where your data lives, and only model updates (such as model weights and parameters), together with the dataset metadata and summary statistics described below, are sent out to improve the shared global model. Your underlying structures, sequences, and individual data points are never transmitted.
CDD also hosts and operates the TuneLab models themselves. When you run predictions, your structures and sequences are processed entirely within CDD Vault and are never sent to Lilly or any other TuneLab participant.
To summarize:
TuneLab membership is opt-in. You must complete an agreement with Lilly before CDD can enable the integration.
You choose which data to contribute. Only data that you explicitly select and approve is used for federated learning, subject to the contribution requirements in your TuneLab agreement. CDD provides a convenient interface for you to select data and carefully review it prior to contribution.
Your raw data does not leave CDD Vault. Your structures and assay results stay in the same CDD Vault environment your organization has already reviewed and approved. They are not transferred to Lilly or other external parties.
Lilly reviews dataset summaries, not raw data. To confirm data quality, Lilly reviews information about assay conditions and protocols, along with aggregated statistics derived from the dataset. This review does not require access to your structures, sequences, or individual data points.
Only model updates are shared. As part of federated learning, model updates are exchanged with the third-party hosted central server to update the global model; your raw data is never shared. Neither Lilly nor any other TuneLab member sees the raw data.
Leaving TuneLab. If you leave TuneLab, the data you contributed for training will be retained or deleted in accordance with the applicable data agreement, and you can request deletion. Note that the influence your data has already had on trained models cannot be removed, and any contribution obligations accrued before you leave continue to apply.
Getting started
Lilly TuneLab is designed for biotech companies wanting to use AI to accelerate their preclinical small molecule or antibody programs. More than 135 member companies have joined TuneLab as of October 2026.
Access requires TuneLab membership and configuration of the integration in CDD Vault. If your organization is not already a TuneLab member, talk to us and we will help you with Lilly’s onboarding process.
Once your organization has access, our support team will enable and configure the CDD Vault integration with TuneLab. After configuration, you will be able to access the models in all CDD Vaults for your account.
How to make predictions
CDD Vault provides access to two TuneLab models: ChemLab and AbLab. ChemLab covers ADMET properties like kinetic solubility, microsomal clearance, or hERG binding. AbLab focuses on antibody developability and predicts properties like melting temperatures or non-specific binding.
Every endpoint prediction returned on the TuneLab platform is accompanied by a statistically rigorous uncertainty estimate: a prediction interval for regression endpoints (e.g. fraction unbound) or a prediction set for classification endpoints (e.g. hERG binding).
You can make and analyze predictions in several places: evaluate a registered molecule or antibody, compare a set of molecules, or use them when exploring ideas. Let’s see how each of these works in detail.
Predict properties for an individual molecule or antibody
Use this workflow to evaluate a candidate while reviewing its record in CDD Vault. Once TuneLab is enabled, you will find a new section on the Molecule/Entity Overview page. This section will contain all predictions you made with the TuneLab models.
To add predictions, click the “Edit” link. This will open the “Select TuneLab models” dialog. The right panel contains all relevant models; here we show the small molecule models, organized into several groups. You can select all models, individual groups, or individual models. If you regularly use a particular set of models, you can save it as a set for quick access. Using the icons next to the set name, the models of a set can be added or removed from the current selection. Existing sets can be changed and removed.
The antibody models only apply to standard monospecific antibodies. The full heavy- and light-chain sequences are required. Bispecific antibodies are currently not supported.
Run bulk predictions for a group of molecules
The bulk prediction functionality provides model predictions for up to 200 molecules. It can be found in the new TuneLab tab of the main “Explore Data” page. After selecting the tab, you can paste a list of identifiers or SMILES strings into the text box. Here, we use a list of antibody identifiers as an example.
For ease of use, you can either select one of your saved sets or use “Select models…” to use the model selection dialog described above. Click “Predict” to start the calculation. For further analysis, download the results as a CSV file.
Explore ideas in the structure editor
The third option uses TuneLab for you to explore ideas in the structure editor. With TuneLab activated, you will find an additional section in the monitor side panel. Click the gear icon to open the model selection dialog and select models that you want to apply to your ideas. Here we selected the fraction unbound in human plasma model.
How to contribute data
For security, only account administrators can contribute data. Account administrators will see an additional “Data Contributions” tab in the TuneLab section.
In the first step, you decide the data category to which you want to contribute and then fill out the category-dependent metadata. You can also upload a PDF that describes your protocol.
Clicking “Next” validates the entry and, if successful, takes you to the next step, where you select the protocol readout that contains the data for contribution. If the protocol has conditions, select the ones that apply.
This step allows you to have a final review of the data contribution before you “Submit to TuneLab”.
This contribution will show up in the data contribution table as “Submitted.” After training, the status will change to “Trained.”
How to check your credit usage
Each prediction consumes credits. It is useful to check how many credits were used and how many are left. Go to the TuneLab / Credit usage tab.
Questions?
Interested in bringing Lilly TuneLab predictions into your CDD Vault? Contact your CDD account manager or reach out to support@collaborativedrug.com, and we'll help set you up. As always, we'd love to hear your feedback.