daphnia_droplets

Statistical analysis pipeline for toxicological experiments using Daphnia, examining the effects of chemical and food treatments on lipid droplet accumulation measured by confocal microscopy.

---

Overview

Daphnia were exposed to chemicals (e.g. TBT) across three food regimes:

Code	Treatment
NF	No food
LF	Low food
HF	High food

After exposure, individuals were imaged by confocal microscopy. Lipid droplets were segmented using deep learning, yielding per-animal summary statistics and individual droplet measurements. This pipeline loads those outputs, fits generalized linear models (GLMs) and linear mixed models (LMMs), and writes diagnostic plots and pairwise comparison tables to disk.

---

Repository structure

daphnia_droplets/
├── bin/
│   ├── analysis_scalable_chems.R     # Main multi-experiment pipeline
│   └── pilot_analysis_only_food.R    # Pilot analysis (food treatment only)
├── data/
│   └── <experiment>/                 # One folder per chemical/experiment
│       ├── measures_export_<N>.csv
│       ├── measures_export_<N>_wells.xlsx
│       └── masks_<N>/
│           └── <well>_droplets.csv   # One file per individual
└── output/
    └── <experiment>/
        ├── graphs/
        └── results/

---

Input data format

All data for an experiment must be placed in data/<experiment_name>/. Experiment folders named pilot (case-insensitive) are excluded from batch runs.

Per-well summary table — measures_export_<N>.csv

One file per replicate (e.g. measures_export_1.csv, measures_export_2.csv). Each row is one imaged well. Must contain at least:

Column	Description
well_name	Well identifier (e.g. A1); first character is the row letter
daphnia_area	Body area (used to normalise droplet count)
droplet_intensity_total	Total lipid-droplet fluorescence intensity
droplet_area_total	Summed area of all droplets
daphnia_size	Body-size proxy
num_droplets	Count of detected droplets

Treatment map — measures_export_<N>_wells.xlsx

One Excel file per replicate, with three columns:

Column	Description
well	Capitalised row letter (e.g. A, B)
food_treatment	"no food", "low food", or "high food"
chem_treatment	Chemical treatment label (can be a single value if no chemical)

Individual droplet data — masks_<N>/<well>_droplets.csv

One CSV per individual (well), inside a masks_<N>/ subfolder for each replicate. Each row is one segmented droplet and must contain an area column. Files are optional; if absent the per-droplet LMM is skipped.

---

Running the analysis

Open R and source the main script:

source("bin/analysis_scalable_chems.R")

A dialog box will appear listing all available experiments. Select one, or choose ALL to process every experiment sequentially.

For the pilot experiment (food treatment only, no chemical), use:

source("bin/pilot_analysis_only_food.R")

Dependencies

The script will install missing packages automatically. Required packages:

tidyverse, readxl, multcomp, lme4, emmeans, DHARMa, MASS, patchwork, car

---

Statistical models

Response variables

Response	GLM family	Description
droplet_intensity_total	Gamma (log)	Total lipid-droplet fluorescence
droplet_area_total	Gaussian	Summed droplet area
daphnia_size	Gaussian	Body-size proxy
num_droplets	Negative binomial	Count of detected droplets
num_droplets_per_area	Gamma (log)	Droplet count ÷ body area

Response variables and their families are defined in the response tibble near the top of analysis_scalable_chems.R and can be modified there.

Model structure

Two GLMs are fitted per response variable:

• Main model — response ~ food_treatment * chem_treatment (or ~ food_treatment if only one chemical level is present)
• Replicate-adjusted model — adds + replicate as a fixed covariate to account for batch effects

Pairwise contrasts are computed with emmeans and p-values are adjusted using the Benjamini–Hochberg (BH) method. Compact letter display (CLD) groups are overlaid on boxplots.

For individual droplet area, a Gamma LMM (glmer) is fitted with well_name as a random intercept to account for pseudo-replication within wells.

Quality control

Wells with five or more zero-valued numeric columns are flagged and removed before modelling (likely imaging failures). A warning listing the affected wells is printed to the console.

---

Output

Results are written to output/<experiment>/.

graphs/

File	Contents
histograms.pdf	Distribution of each response variable, overall and by treatment group
<response>_by_treatment.pdf	Boxplot with jittered points (coloured by replicate) and CLD letters
individual_droplets_by_treatment.pdf	Violin plot of individual droplet area by treatment
model_diagnostics.pdf	DHARMa residual diagnostics for each GLM
individual_droplet_area_diagnostics.pdf	Diagnostics for the droplet-area LMM

results/

File	Contents
<response>_anova_treatment.csv	Type II ANOVA table for the main model
<response>_anova_replicates.csv	Type II ANOVA table for the replicate-adjusted model
<response>_pairs_treatment.csv	BH-adjusted pairwise contrasts (main model)
<response>_pairs_replicates.csv	BH-adjusted pairwise contrasts (replicate-adjusted model)
individual_droplet_area_pairs_treatment.csv	Pairwise contrasts for droplet area LMM

---

Example data layout

The data/TBT-CL/ folder illustrates the expected structure for a two-replicate experiment:

data/TBT-CL/
├── measures_export_1.csv
├── measures_export_1_wells.xlsx
├── measures_export_2.csv
├── measures_export_2_wells.xlsx
├── masks_1/
│   └── *.csv   (one file per individual)
└── masks_2/
    └── *.csv