Zebrafish Obesogenic Test, label‑free

ViZOT quantifies anal fin ray and ventral subcutaneous adipocytes directly on the brightfield image acquired at each ZOT session, in one step: no crop, no background subtraction, no threshold to tune.

Label-free Adiposity & ZOT Methodology

Obesogens promote fat storage, and some of them induce a thrifty phenotype in which adiposity is preserved despite fasting. The ZOT identifies such compounds in vivo by measuring, in the same larva, the area of subcutaneous adipocytes before and after a 24-h exposure in a fasting state. Staining every larva twice with Nile Red adds a reagent that may interact with the tested compounds and limits repeated measurements on the same animal.

01

Whole-Organism ZOT Assay

Larvae of 8.5 to 9.5 mm standard length are imaged before and after a 24-h exposure in a fasting state, and each larva is its own reference for the adiposity variation.

In Vivo AssayThrifty Phenotype
02

Brightfield Acquisition

Adipocyte lipid droplets appear as refractile spheres at the 4× magnification of the ZOT. The brightfield image is already the first channel of every ZOT acquisition.

4× Objective2048 × 2048
03

One-Step U-Net

A single U-Net with a ResNet-34 encoder segments AFRSAT and VSAT adipocytes on the whole image. ViZOT-BF reads the brightfield image, ViZOT-NR the Nile Red fluorescence image.

Deep LearningBF / Nile Red
04

ZOTAS-Compatible Readout

Adipocyte areas in pixels and mm², area variation normalized to the controls and statistics versus the control, in the folder layout and Excel tables of ZOTAS.

AFRSAT / VSATBatch Analysis

Same conclusions as ZOTAS

The larvae of the published screen were reanalyzed with ViZOT-NR and ViZOT-BF and compared, larva by larva, with the published ZOTAS measurements.

12

Compounds, 28 Experiments

486 larvae for AFRSAT and 500 for VSAT, analyzed with the three methods on identical sets of animals.

Published Screen
4/4

Obesogens Identified

Amiodarone, rosiglitazone, triclosan and dibutyl phthalate induced a thrifty phenotype with the three methods (p < 0.0001); the eight other compounds showed no significant difference.

AFRSATVSAT
2/2

Unseen Compounds Classified

Amiodarone (obesogenic) and DEHP (non-obesogenic), withheld from training and validation, were correctly classified by both models.

Held-out Test
24

Excluded Larvae Recovered

All the larvae with images that were left out of the published analyses (ZOTAS threshold failure, manual exclusion) were quantified, in one step and without operator intervention.

No Threshold
Peer-Reviewed Literature & Protocols

Associated Scientific Publications

The ZOT bioassay, its optimized protocol and the ZOTAS software, and the 2026 screen whose larvae were reanalyzed with ViZOT, which identified the same obesogens as ZOTAS.

Area variations measured with ViZOT were correlated with those of ZOTAS (Spearman r = 0.98 and 0.96 for ViZOT-NR, 0.88 and 0.86 for ViZOT-BF, AFRSAT and VSAT), areas on single images even more closely (r ≥ 0.98, slopes 0.95 to 1.03), and the dispersion of the area variation among control larvae was similar with the three methods.

How ViZOT Works

Processing plots written by the desktop application for one larva: input image, ViZOT probability map and the adipocytes it measures.

ViZOT-BF processing plot of the AFRSAT
01

Brightfield AFRSAT

ViZOT-BF finds the anal fin ray adipocytes on the whole brightfield frame, without crop or threshold selection.

ViZOT-BF processing plot of the VSAT
02

Brightfield VSAT

The same network measures the ventral adipocytes: one image per depot, areas in pixels and mm².

ViZOT-NR processing plot of the AFRSAT
03

Nile Red

ViZOT-NR reads the Nile Red fluorescence image of the same acquisition, for experiments analyzed with both channels.

Desktop Application
ViZOT Logo

Download ViZOT

The desktop application runs the ViZOT-BF and ViZOT-NR networks on whole ZOT experiments, offline: one installer for every Windows PC, using the NVIDIA GPU when it is faster than the processor.

Downloads are currently restricted to authorized accounts. After publication, ViZOT will be available to registered users once their registration has been validated.

ViZOT processing plot preview
ViZOT v0.2.0
Windows
System requirements
  • Windows 10/11 x64
  • >8 GB RAM
  • NVIDIA GPU optional
Microscopy requirements
  • .tif, brightfield first channel
  • 2048 × 2048, 4× objective

ViZOT was trained on images from one set-up (Nikon Eclipse Ni, 4× Plan Apo, ORCA-Flash4.0); other microscopes and cameras remain to be evaluated.

AI Source Code

Explore ViZOT-BF & ViZOT-NR

The training and inference code and the weights of both models will be released with the article.

ViZOTTraining, inference & weights
On publication

For help adapting ViZOT to your own microscope, contact theo.merce@danio-tracker.com .

Credits & Institutional Partners