VCDD / RESEARCH MATERIALS & DOCUMENTATION

Resources

Paper, code and data releases, downloadable molecular records, and documentation for working with VCDD.

RESEARCH RELEASES

Paper, code & data

The VCDD manuscript, research implementation and associated datasets will be available here as they are released.

VCDD paper

Manuscript and supplementary information.

PDF / SUPPLEMENTARY MATERIALS

Release pending

Research code

Model implementation, training and evaluation workflows, with reproducibility instructions.

SOURCE CODE / ENVIRONMENT / DOCUMENTATION

Release pending

Research data

Processed research datasets, source-data records and dataset documentation.

DATA ARCHIVE / METADATA / PROVENANCE

Release pending

Full research releases are not yet linked. Molecular records and the archived computational trajectory are already available in Available downloads.

Usage guide

Open the workspace. Select VCDD in the top navigation. The design studio opens over the current page without an intermediate page.
Inspect the workflow. Target, generation, dosing, virtual evaluation, feedback, screening and human review form the seven stages. Stage controls and playback are available in the studio. Pausing the visualization does not pause a server job.
Inspect a molecule. Choose a VCDD record to view its identity, paired condition and archived concordance values. Switch between its 2D structure and accepted rotatable 3D display conformer.
Examine the evidence. Research contains the ten Figure 6 molecular records and the separate Figure 5j PC3 computational replay. Conditions and readout sources remain attached to each result.
Prepare a future model run. The model-run configuration requires a connected model service and a verified target catalog. New generation and job submission remain unavailable until that connection is established.

Frequently asked questions

Does opening the design studio run the trained models?

No. This version opens the interactive workspace and archived VCDD records. The model service is not connected, so no new molecular generation, model evaluation or GPU job is submitted.

Are the molecular structures real VCDD outputs?

The ten Figure 6 structures and the Figure 5j PC3 pool come from archived model-generated records with retained identities and provenance. They are not newly generated by visiting the website and do not establish synthesized material or measured activity.

Does the dosing animation predict the cell's physical response?

No. Dosing and cell magnification are a process schematic. The displayed cell is not a dynamic biological simulation, a growth measurement or an observed drug-response image.

Why is VCDD-05 available only in 2D?

No new display conformer passed coordinate-derived stereochemical identity verification. Its original molecular record is preserved; rejected coordinates are not substituted or presented as an accepted 3D structure.

Why are VCAP, HT29 and PC3 records shown separately?

Figure 6 contains nine VCAP designs and one HT29 design under their own doses. Figure 5j is a different PC3 condition. The website does not combine these into a single trajectory or transfer one condition's readouts to another.

Can the generated molecules be sent directly for synthesis?

Not on the basis of these predictions alone. Route feasibility, exact material identity, specified stereochemistry, analytical purity, solubility and assay conditions need separate review. No procurement clearance is issued by this website.

Can I request twenty molecules for a new cellular target now?

That requires a connected, validated model service and an explicitly defined generation and screening budget. The current website does not produce new candidates or relax screening rules to fill a requested count.

File formats & field definitions

Downloads retain source identities and exact treatment conditions. A display ID never replaces the original archive key.

Format	Contents	Current availability
CSV / SMILES	One molecule per row; stereochemical SMILES, identifiers, cell, dose, exposure and concordance values.	Ten Figure 6 records
JSON / molecular evidence	State vectors, source percentiles, molecular properties, provenance hashes and accepted display conformers.	Ten Figure 6 records
JSON / query trajectory	128 fixed candidates, eight ordered queries, source-specific predictions and final recommendation.	Figure 5j PC3 record
SDF	Structure connection table and optional coordinates. Display geometry must not be treated as original model-input geometry.	Not included in the current downloads

Identity & condition fields

Field	Meaning
`id` / `source_id`	VCDD presentation identifier / immutable source record identifier.
`smiles` / `inchikey`	Archived stereochemical structure string / molecular identity key.
`cell_id`	The cellular context paired with that record, not a general efficacy indication.
`dose_um` / `time_h`	Source dose in micromolar / exposure in hours. Values are not rescaled to improve a score.
`gene_r` / `pathway_r`	Target-prediction Pearson correlations over 978 genes / 101 pathways.
`vectors`	Labels and target/predicted arrays in their archived order. These are not a generic upload specification for a different model.
`conformer.status`	Whether an accepted display conformer is available. Unavailable geometry must not be inferred or replaced silently.

Available downloads

VCDD-01 to VCDD-10SMILES, identities, exact conditions and archived concordance values · CSV

CSV

Full molecular evidence978-gene and 101-pathway vectors, source percentiles, provenance hashes and display coordinates · JSON

JSON

Recorded feedback trajectoryFigure 5j · PC3 · 128 candidates · eight queries · JSON

JSON

VCDD usage & data guideWorkflow, scientific limits, field definitions and available formats · Markdown

Guide

Figure 6 contains nine VCAP records and one HT29 record. The Figure 5j PC3 trajectory is a different condition and is not combined with those results. VCDD-05 is 2D-only: no display conformer passed the coordinate-derived stereochemical identity check.

What each readout means

Gene-state concordance: Pearson correlation between target and predicted expression across 978 genes. Pathway concordance: correlation across 101 pathway-level values. Both are computational agreement metrics, not experimental drug efficacy.

Sensitivity percentile: the rank of the source-specific KillRead prediction within its archived candidate pool. KillRead uses fixed dose-integrated features. It is not IC50, a single-dose death probability or a measured viability value. GDSC1+2 includes GDSC2; these readouts are not independent replicates.

QED / SA: drug-likeness and synthetic-accessibility heuristics. A favorable score does not establish a viable synthesis route, purity, stability or safety.

Before experimental handoff

Initial chemical feasibility and a favorable virtual-cell result do not complete wet-lab readiness review. Structure identity, specified stereochemistry, conformer geometry, model support and assay conditions require a separate evidence trail.

Consultation with a synthesis provider must confirm route feasibility, material identity and purity. Solubility, solvent tolerance, dose delivery and reproducible cellular activity require physical measurements. No record on this site is released for procurement.

Model service

The design interface is available now; live inference is not connected. No GPU job, new molecular generation or strict-screening calculation runs in this local version.

A model server must supply a compatible protocol declaration, verified target catalog and real job results. Weights stay on the inference server, not in the public website files.

Model-run configuration