Q1 2026 Product Update Webinar Recap: Zero-Click Inference and Expanded Registration Forms

CDD's Q1 2026 product update webinar covered two feature areas. The first, zero-click inference, automates the construction and deployment of predictive structure-activity models within CDD Vault. The second, registration form improvements, expand structured data capture for small molecules, siRNA, and antibody-drug conjugates (ADCs). Both capabilities address operational inefficiencies that accumulate when predictive modeling and entity registration are handled outside the primary data environment.

Zero-Click Inference

Building predictive QSAR models has traditionally required researchers to export experimental data, prepare descriptors, train models externally, and redeploy predictions back into their working environment. Zero-click inference removes that overhead entirely.

How it works:

• CDD Vault automatically trains predictive models in the background using dose-response data already stored in the system
• Model training, quality control, and deployment require no researcher configuration
• Models update automatically each time new protocol data is added to the vault
• A minimum of 6 data points is required to initiate training; 30 or more data points are recommended for reliable predictions
• All models pass a five-fold cross-validation and must meet an R-squared threshold greater than 0.4 before deployment
• Models are released in ONNX format and execute locally in the browser — no data is transmitted outside the vault

Where predictions appear in the vault:

• On the molecule page, below calculated properties
• In the structure editor, updating in real time as structures are modified
• Alongside bioisostere suggestions (requires the AI module)
• Alongside deep-learning similarity results, including commercially available compounds from Enamine and patent-novelty results from SureChEMBL

Protocol conditions such as gene name allow separate models to be maintained for multiple targets within a single dose-response protocol. Each model is independently trained and displayed per compound record with a point estimate and confidence interval.

Researchers working with virtual compounds, including those generated through Boltz-2 or DiffDock docking workflows, can register idea structures under a distinct prefix and review inference predictions alongside docking scores prior to synthesis.

A manuscript describing the underlying methodology is available through CDD for teams requiring documentation for internal scientific review.

Registration Forms

Prior to this release, a single vault supported one registration system applied uniformly across all entity types. Registration form improvements allow multiple registration systems to coexist within the same vault, with each form defining the entity-level, batch-level, and sample-level fields relevant to a specific molecule type.

Key capabilities introduced:

• Multiple registration systems within a single vault, each with a configurable prefix and starting compound number
• Entity-specific fields so that each modality displays only the metadata fields applicable to that type
• Support for both auto-generated and custom naming schemes at the molecule and batch level
• A batch link data type that associates records across entity types during bulk import

Supported entity types demonstrated:

• Small molecules — standard structure-based registration with configurable fields for stereochemistry notes, scaffold series, and vendor identifiers
• siRNA — sequence-based registration via the macromolecule composer, capturing sense and antisense strands, modification patterns, base pairing relationships, and custom nucleotide monomers; produces chemically aware records supporting hairpins, dumbbells, conjugates, and strand breaks
• Antibody-drug conjugates (ADCs) — registers heavy and light chains, variable and constant domain annotations, disulfide bridge locations, and conjugation point information for linker and payload; payload molecules can be called from existing registered records; supports partial information capture for programs where conjugation site data is not yet fully characterized

Iterative biologic variants, such as antibody sequences with single amino acid substitutions, can be tracked using a parent entity ID in a molecule-level metadata field. A keyword search on that ID returns the parent record and all variants, which can be compared using the customizable report view or the visualization tool with assay data overlaid.

Registration forms are approaching general release. Vault administrators will receive email notification at deployment, and a banner will appear at login when the feature goes live.