Response across concentrations
Green bars show between-day training variability, not confidence or prediction intervals. The connecting line joins observed concentration levels only. No curve is fitted between them.

Explore a published cell assay. Hold back an entire experimental day. See what a simple baseline predicts, where it fails, and what the evidence can support.
Green bars show between-day training variability, not confidence or prediction intervals. The connecting line joins observed concentration levels only. No curve is fitted between them.
Can the average response at each concentration, learned from five experimental days, describe a sixth day? The answer depends on which day is withheld. Variation in raw readout levels can dominate performance.
This is a transparent reference benchmark. It is not a deep learning model, a mechanistic cell simulation, or an assessment of treatment efficacy.
Each fold trains on the other five days. Lower MAE / RMSE is better. Negative R² means predictions perform worse than the held-out day's own mean. The global baseline uses the training mean, never the test mean.
Each point is one held-out measurement from the selected day. The diagonal is perfect agreement.
All test concentrations have matches in the corresponding training fold. No interpolation, normalisation, hyperparameter search, test-data fitting or extrapolation is performed. Experiment identifiers are labels, not decoded calendar dates.
All source rows retained. No missing values are imputed. Row identifiers preserve the source CSV's rownames field.
The report includes the selected split, both baselines, fold metrics, row-level predictions, provenance and limitations. It contains only calculations from this public dataset.
Dataset: drc::vinclozolin, redistributed as CSV by Rdatasets. 53 observations across six experiment identifiers. Package documentation identifies a Chinese hamster ovary AR reporter assay and 22-hour exposure.
Original study: Nellemann, Dalgaard, Lam & Vinggaard (2003), Toxicological Sciences 71, 251–262.
Package documentation ↗
Find original study ↗
Provenance manifest ↗
Bundled source help ↗
CSV retrieved 5 November 2026. SHA-256:
a7de299910ae8c8296ee513b6e959b09f76821e1cfe3d755236b545ee8194c9f
Package licence: GPL-2 or GPL-3. Attribution and source help retained. Explorer code is GPL-3.0-or-later. Licence text.