Real photographs: the complete 14-day workflow

REAL DATA · EIGHT ORIGINAL PHOTOGRAPHS · TWO MATCHING SETS

From the original photograph to a traceable comparison.

A complete practice run with supplied metadata, five spots per target colony, image QC, matched coefficients, saved tables and publication-format exports.

Start here after installing FunMel. For your own photographs, use the Get started workflow; this article uses eight fixed example filenames and their supplied design. Run numbered steps in order; all objects used below are created in this article. The figures and tables shown here use the bundled photographs. The function catalog covers specialist tools separately.

What this experiment contains

Each matching set has one independent control plate and three independent confrontation plates. Each plate has four target Zymoseptoria tritici colonies. Confrontation plates also contain a central fungus and connected hyphal growth; control plates do not. The central growth is not an inhibition zone.

Matching set	Independent plate image	Role	Replicate
A60_2_14_days	2026_10_02_99_38.JPG	A60_2 confrontation	1
A60_2_14_days	2026_10_02_99_39.JPG	A60_2 confrontation	2
A60_2_14_days	2026_10_02_99_40.JPG	A60_2 confrontation	3
A60_2_14_days	2026_10_02_99_41.JPG	Own-set control	1
B20_2_14_days	2026_10_02_99_42.JPG	B20_2 confrontation	1
B20_2_14_days	2026_10_02_99_43.JPG	B20_2 confrontation	2
B20_2_14_days	2026_10_02_99_44.JPG	B20_2 confrontation	3
B20_2_14_days	2026_10_02_99_45.JPG	Own-set control	1

The folder-derived labels are teaching labels whose biological identities and assignments must be checked against laboratory records. The supplied source field fungus="hm83" is retained without assigning an unverified biological role. Unrecorded strain/medium fields remain blank. The eight original JPEGs are unchanged. No separate image download is required.

Units matter. This is eight plates, 32 target colonies and, when all requested spots fit, 160 spots. It is not 160 independent biological replicates. One control per set supports descriptive coefficients, not the package’s replicated-cell significance tests.

1. Locate the photographs and create one practice project

library(FunMel)
photos14 <- system.file("extdata","14-days",package="FunMel")
stopifnot(nzchar(photos14))
project_dir14 <- tempfile("FunMel_14_day_project_")
project14 <- prepare_experiment(photos14,project_dir14)

tempfile() supplies a new temporary folder name, so this teaching run does not overwrite earlier work. Its files last only as long as the temporary location remains. For a project you want to keep, replace that assignment before starting with a new durable folder, for example:

# Alternative to tempfile(): keep the bundled practice project in your home folder.
project_dir14 <- file.path(path.expand("~"), "FunMel_14_day_project")
if (file.exists(file.path(project_dir14, "project.rds"))) {
  project14 <- load_experiment(project_dir14)
} else {
  project14 <- prepare_experiment(input=photos14, project_dir=project_dir14)
}

project_dir14 is the output folder, not the input ZIP. To analyze your own ZIP, use the separate input/output example in Get started.

Do not run both preparation alternatives for the same project. To revisit work already saved, go directly to step 8. A folder or ZIP is passed directly here; there is no file chooser and no opportunity to select a QC image accidentally.

project14$project_dir   # All output files.
project14$metadata_file # Project copy of the experimental design.
project14$qc_dir        # QC PNGs created by the image run.

2. Check the supplied metadata

The combined metadata is discovered automatically. Inspect it before measuring:

metadata14 <- read.csv(project14$metadata_file,check.names=FALSE)
knitr::kable(metadata14[c("plate_id","set_id","treatment","centre_model","replicate")],
             row.names=FALSE)

plate_id	set_id	treatment	centre_model	replicate
2026_10_02_99_38	A60_2_14_days	A60_2	fungus	1
2026_10_02_99_39	A60_2_14_days	A60_2	fungus	2
2026_10_02_99_40	A60_2_14_days	A60_2	fungus	3
2026_10_02_99_41	A60_2_14_days	control	none	1
2026_10_02_99_42	B20_2_14_days	B20_2	fungus	1
2026_10_02_99_43	B20_2_14_days	B20_2	fungus	2
2026_10_02_99_44	B20_2_14_days	B20_2	fungus	3
2026_10_02_99_45	B20_2_14_days	control	none	1

Each control appears once, labelled treatment="control" and centre_model="none". Each confrontation uses centre_model="fungus". All four plates in a matching group share their set_id. A60_2 never uses B20_2’s control or vice versa. replicate is numbered within a treatment and set; it does not by itself pair a treatment plate with a uniquely replicated control. With one control in a set, its mean is the shared denominator for the three treatment plates.

Verify treatment identities, controls, sets, day and any recorded strain/medium. Correct missing or inaccurate information. Additional user-defined columns are preserved. Do not infer biological identities from a filename alone. If corrections are needed, edit the project CSV and save it before matched comparisons. Do not change image identity or layout after analysis without remeasuring.

3. Set calibration and measure all eight plates

In this example, 90 is an illustrative outer dish diameter, not a verified physical measurement of these photographs. Replace it with the actual outer diameter for scientific use. Do not paste subsequent R commands into a readline() prompt; this workflow sets the number directly.

dish_mm <- 90
result14 <- run_experiment(
  project14,dish_diameter_mm=dish_mm,max_dim=1200,
  n_spots=5,spot_radius_mm=0.4,n_boot=20
)
result14$analysis_status
#> [1] "complete"
result14$failed
#> [1] file  error
#> <0 rows> (or 0-length row.names)
stopifnot(nrow(result14$failed)==0L)

The expected status is "complete"; the failure table has zero rows. This saves image_analysis.rds, measurement CSVs and a QC PNG for each plate. A genuine failed image must be corrected before finalization. Success does not certify segmentation or biological accuracy.

Setting	Meaning in this tutorial
dish_diameter_mm	Outer diameter used to convert pixels to millimetres
max_dim=1200	Longest working-image dimension; influences spatial resolution
n_spots=5	Request five additional circular regions within each target colony
spot_radius_mm=0.4	Radius of each spot, not its diameter
n_boot=20	Quick teaching setting for spatial-block bootstrap summaries, not precise uncertainty

Whole-colony MI uses eligible interior pixels, excluding the rim and saturated/specular pixels. Spots add a deterministic, spatially spread sample; they do not replace whole-colony measurements. Omit spot settings for whole-colony measurements alone. Use consistent settings across comparable plates.

data.frame(plates=nrow(result14$plates),colonies=nrow(result14$colonies),
           spots=nrow(result14$spots),failed=nrow(result14$failed))
#>   plates colonies spots failed
#> 1      8       32   160      0

4. Inspect masks, spots and illumination

Inspect all eight QC images in project14$qc_dir. The examples below show one confrontation and its own control; they do not replace reviewing the remaining photographs.

Example QC: a confrontation plate and its own matching control. Inspect the target outlines, spot locations, central fungal tissue/hyphae and dish calibration on every image.

Example QC: a confrontation plate and its own matching control. Inspect the target outlines, spot locations, central fungal tissue/hyphae and dish calibration on every image.

Check target-colony boundaries, central tissue/hyphal separation, spot containment, dish edge, shadows, glare and warning text. The approximately round targets need not have perfectly circular outlines. Correct unsuitable images or measurements before interpreting the final comparisons.

fields14 <- c("plate_id","n_colonies_detected","central_detected","flat_field_status")
qc_table14 <- result14$plates[intersect(fields14,names(result14$plates))]
missing14 <- setdiff(fields14,names(result14$plates))
if(length(missing14)) cat("Not reported: ",paste(missing14,collapse=", "),
                        "; no values were inferred. Inspect QC before interpretation.\n",sep="")
knitr::kable(qc_table14,row.names=FALSE)

plate_id	n_colonies_detected	central_detected	flat_field_status
2026_10_02_99_38	4	TRUE	withheld_degenerate_fit
2026_10_02_99_39	4	TRUE	withheld_degenerate_fit
2026_10_02_99_40	4	TRUE	withheld_degenerate_fit
2026_10_02_99_41	4	FALSE	withheld_degenerate_fit
2026_10_02_99_42	4	TRUE	withheld_degenerate_fit
2026_10_02_99_43	4	TRUE	withheld_degenerate_fit
2026_10_02_99_44	4	TRUE	withheld_degenerate_fit
2026_10_02_99_45	4	FALSE	withheld_degenerate_fit

Read the status recorded in your table. withheld_degenerate_fit means the fitted illumination correction did not pass the saturation safeguard and was withheld. NA or an absent field means not reported, not that correction was applied or passed. The table explicitly identifies fields not supplied by the analysis. Inspect the QC images and repeat measurement if QC information is missing. A withheld correction is not evidence of adequate illumination. Treatment/control ratios do not guarantee cancellation of exposure or lighting differences. Use consistent imaging and a suitable reference; rephotograph when needed. The figures here illustrate software output, not a validated melanin-concentration assay.

Console reminders identify image-QC checks. Inspect the QC images before interpreting or reporting results.

5. Inspect whole-colony, spot and profile tables

fields14 <- c("plate_id","colony_id","MI_min","MI_mean","MI_max","MI_spot_mean","n_spots_measured","spot_status")
colony_table14 <- result14$colonies[intersect(fields14,names(result14$colonies))]
knitr::kable(head(colony_table14),digits=2,row.names=FALSE)

plate_id	colony_id	MI_min	MI_mean	MI_max	MI_spot_mean	n_spots_measured	spot_status
2026_10_02_99_38	TL	42.26	70.45	90.67	72.00	5	ok
2026_10_02_99_38	TR	41.95	72.11	95.23	71.60	5	ok
2026_10_02_99_38	BR	53.71	74.29	94.55	71.19	5	ok
2026_10_02_99_38	BL	43.64	69.64	88.77	61.21	5	ok
2026_10_02_99_39	TL	40.52	71.19	93.22	71.04	5	ok
2026_10_02_99_39	TR	34.74	68.28	93.39	64.63	5	ok

MI = 100 - CIELAB L*; larger values indicate darker tissue. MI_min and MI_max are observed whole-colony pixel extrema, not confidence limits. MI_spot_mean is the mean of the individual spot means. The corresponding MI_spot_min/max are extrema of spot means, not pixel extrema.

fields14 <- c("plate_id","colony_id","spot_id","spot_radius_mm","n_pixels","MI_mean","MI_min","MI_max")
spot_table14 <- result14$spots[intersect(fields14,names(result14$spots))]
knitr::kable(head(spot_table14),digits=2,row.names=FALSE)

plate_id	colony_id	spot_id	spot_radius_mm	n_pixels	MI_mean	MI_min	MI_max
2026_10_02_99_38	TL	1	0.4	37	71.71	63.31	85.44
2026_10_02_99_38	TL	2	0.4	37	74.01	65.28	87.13
2026_10_02_99_38	TL	3	0.4	37	69.63	55.06	80.29
2026_10_02_99_38	TL	4	0.4	37	68.49	59.81	78.25
2026_10_02_99_38	TL	5	0.4	37	76.13	66.56	84.13
2026_10_02_99_38	TR	1	0.4	37	71.61	63.28	81.61

Each spot has a location in working-image pixels (x_px, y_px), a physical radius and an achieved pixel count. L_mean is mean CIELAB lightness; spot MI_mean is 100-L_mean. Pixel counts depend on the scale and working resolution. Check spot_status and n_spots_measured; if all requested spots cannot fit without overlap, the incomplete summary is unavailable instead of invented.

fields14 <- c("plate_id","colony_id","mid_mm","MI_mean","MI_lo","MI_hi")
profile_table14 <- result14$profiles[intersect(fields14,names(result14$profiles))]
knitr::kable(head(profile_table14),digits=2,row.names=FALSE)

plate_id	colony_id	mid_mm	MI_mean	MI_lo	MI_hi
2026_10_02_99_38	TL	0.25	71.19	70.28	71.88
2026_10_02_99_38	TL	0.75	72.96	71.71	73.82
2026_10_02_99_38	TL	1.25	68.91	67.48	70.57
2026_10_02_99_38	TL	1.75	68.08	66.74	69.64
2026_10_02_99_38	TL	2.25	68.01	66.21	69.70
2026_10_02_99_38	TL	2.75	69.58	66.31	71.97

Profiles describe darkness from the colony edge inward. Their spatial bootstrap bounds quantify an image-level sampling procedure, not biological replication or chemical melanin uncertainty. The low bootstrap count here is for speed.

The full metadata is in result14$metadata; custom column names and values are retained. Colliding measurement names receive a metadata_ prefix only in measurement tables. Do not use project14$result_file$spots: result_file is a filename, whereas result14$spots is a data frame.

6. Finalize and verify the own-control arithmetic

reviewed14 <- FunMel::finish_experiment(project14)
fields14 <- c("set_id","treatment","MI_treatment_mean","MI_control_mean",
              "melanization_coefficient","n_treatment_plates","n_control_plates")
coefficient_table14 <- reviewed14$melanization_coefficient[
  intersect(fields14,names(reviewed14$melanization_coefficient))]
print(knitr::kable(coefficient_table14,digits=3,row.names=FALSE))

set_id	treatment	MI_treatment_mean	MI_control_mean	melanization_coefficient	n_treatment_plates	n_control_plates
A60_2_14_days	A60_2	68.996	42.374	1.628	3	1
B20_2_14_days	B20_2	67.463	51.813	1.302	3	1

reviewed14$assessment$status

[1] “insufficient_plate_replication”

A plate’s MI is the equally weighted mean of its target-colony MI_mean values. Within each set, the coefficient is:

$$\mathrm{MC}_{s,t}=\frac{\operatorname{mean}(\mathrm{MI}_{\text{treatment plates},s,t})} {\operatorname{mean}(\mathrm{MI}_{\text{control plates},s})}.$$

For this teaching calibration, A60_2 is approximately 1.628 and B20_2 approximately 1.302. Verify the saved arithmetic without pooling controls:

cf14 <- reviewed14$melanization_coefficient
stopifnot(isTRUE(all.equal(cf14$melanization_coefficient,
                           cf14$MI_treatment_mean/cf14$MI_control_mean)))
knitr::kable(reviewed14$per_plate_coefficients,digits=3,row.names=FALSE)

plate_id	treatment	set_id	MI_treatment_plate	MI_matching_control	melanization_coefficient	n_control_plates	qc_review_status	status
2026_10_02_99_38	A60_2	A60_2_14_days	71.621	42.374	1.690	1	unconfirmed	ok
2026_10_02_99_39	A60_2	A60_2_14_days	71.938	42.374	1.698	1	unconfirmed	ok
2026_10_02_99_40	A60_2	A60_2_14_days	63.428	42.374	1.497	1	unconfirmed	ok
2026_10_02_99_42	B20_2	B20_2_14_days	65.057	51.813	1.256	1	unconfirmed	ok
2026_10_02_99_43	B20_2	B20_2_14_days	68.045	51.813	1.313	1	unconfirmed	ok
2026_10_02_99_44	B20_2	B20_2_14_days	69.287	51.813	1.337	1	unconfirmed	ok

The status "insufficient_plate_replication" is expected. There is one control per set, so significance tests are withheld, while valid descriptive ratios remain available. Four colonies and five spots per colony do not add independent control plates. These two treatments also occur in different sets, so a direct causal treatment ranking may be confounded with batch. A darker image-based mean is not a demonstrated fold change in melanin concentration.

For a prespecified spot-mean descriptive response:

spot_coefficient14 <- melanization_coefficient(reviewed14,response="MI_spot_mean")

This changes the response definition; it does not create independent spot replicates. The default figure and assessment use whole-colony MI.

7. Display, export and explain the final figure

print(plot_treatment_comparison(reviewed14))

Descriptive real-data comparison. Each row uses that treatment’s own set control; these are image-darkness ratios under the illustrative calibration.

For your own project, a blank plotting result can mean that matching assignments are incomplete. Check reviewed14$comparison_status and reviewed14$coefficient_error. When the status is awaiting_design, plot_treatment_comparison() reports why it cannot create a matched comparison; use the set-only plot later in this article for descriptive measurements. After actual matching assignments are supplied, finalize again before plotting. Do not replace missing labels with guessed controls. A plot stored in a variable must be explicitly printed in a script or loop.

You can also plot the finalized object already in memory:

print(plot_treatment_comparison(reviewed14))

• Left panel: independent plate MI means. Squares are controls; coloured circles are treatment plates; lower diamonds are treatment means.
• Right panel: each treatment plate divided by its own set’s control mean, and the set-level ratio of means. The dashed reference line is MC = 1.
• Bars and printed ± values: one between-plate SD, not confidence intervals. Control SD is unavailable for a single control. Ratio SD conditions on the shared control mean and does not include its uncertainty.
• QC reminders: remain visible in the console and separate caption file; they identify checks to complete before reporting results.

Finalization already exported treatment_comparison.pdf, .svg, .png and _statistics.csv into the project folder. To create a named alternative:

plot_treatment_comparison(reviewed14,
  file=file.path(project14$project_dir,"my_14_day_comparison"),
  labels=c(A60_2="Fungal treatment A60_2",B20_2="Fungal treatment B20_2"))

PDF and SVG retain vector text and geometry; PNG is 600 dpi. Check the target journal’s final dimensions, font sizes and caption requirements. Changing display labels does not change metadata or statistical matching.

8. Close R, reopen the project and read saved tables

For a project saved in a durable folder, start a fresh session and select its image_analysis.rds file. Read saved results directly from that same folder:

library(FunMel)
saved_file14 <- file.choose() # Select image_analysis.rds in the project folder.
saved_dir14 <- dirname(normalizePath(saved_file14,mustWork=TRUE))
result14 <- readRDS(file.path(saved_dir14,"image_analysis.rds"))
reviewed14 <- readRDS(file.path(saved_dir14,"reviewed_analysis.rds"))
head(result14$spots)
print(plot_treatment_comparison(reviewed14))

This reads existing measurements; it does not repeat image analysis. Keep the complete project folder so the original and finalized results remain together.

For project-aware recovery, the optional load_experiment() and read_experiment() helpers reconnect moved project paths:

project14 <- load_experiment() # Select project.rds in its current folder.
result14 <- read_experiment(project14)
reviewed14 <- read_experiment(project14,finalized=TRUE)

After opening the project with load_experiment(), update final outputs after correcting valid experimental annotations:

reviewed14 <- FunMel::finish_experiment(project14)

Final outputs saved means this succeeded. It is unnecessary to prepare the photographs again. Keep the complete folder, including project.rds, metadata CSV and saved results. load_experiment() reconnects moved internal paths. Saved measurements can be read/finalized without the original photographs, but a new image run still needs them. Missing metadata cannot be safely guessed or rebuilt from image appearance. Changes affecting masks require reanalysis.

9. Optional close-up measurements and specialist tools

Batch results keep tables and saved QC PNGs but drop large image arrays. For interactive image-level views, reprocess just one real photograph with images retained:

row14 <- metadata14[metadata14$plate_id=="2026_10_02_99_38",,drop=FALSE]
plate14 <- analyze_plate(file.path(project14$image_root,row14$relative_file),
  id=row14$plate_id,layout=layout_by_treatment(row14),dish_diameter_mm=dish_mm,
  max_dim=1200,n_spots=5,spot_radius_mm=0.4,n_boot=20,keep_images=TRUE)
plot_qc(plate14)
print(plot_profiles(plate14))
plot_melanization_map(plate14,type="2d")

The darkness map is not pigment concentration. For a different spot count/radius, sample_colony_spots(plate14,n_spots=3,spot_radius_mm=0.4) demonstrates the sampling API; do not change compared responses inconsistently. Calibration, classifier training and interactive annotation examples are in the complete function catalog. A classifier trained on synthetic teaching labels is not validated for these photographs.

Find the exported files

File	Contents
project.rds	Saved project paths for project-aware recovery
plate_metadata.csv	Editable experimental assignments, one row per plate
image_analysis.rds	Original measured tables and analysis status
colonies.csv, spots.csv, profiles.csv, plates.csv	Measurement tables
failed.csv	Failed-image reports; zero rows on a successful run
metadata_used.csv	Full metadata attached to the outputs
reviewed_analysis.rds	Finalized result; the name does not certify review
assessment.rds	Statistical/descriptive assessment and its status
melanization_coefficient.csv	Own-set ratios of treatment/control means
per_plate_coefficients.csv	Treatment plate ratios using own-set control means
coefficient_plate_means.csv	Independent plate means used in calculations
treatment_comparison.pdf/.svg/.png	Comparison exports
treatment_comparison_statistics.csv	Exact descriptive values plotted
treatment_comparison_caption.txt	Full interpretation, SD definitions, limitations and QC status
qc_review_status.csv	Recorded QC status
qc/	One QC PNG per image
sessionInfo.txt	R, platform and package versions

Different recorded day/strain/medium contexts create separate final output folders and context_index.csv; do not pool incompatible contexts. For your own photographs, follow the Get started workflow with your folder or ZIP and correct segmentation layouts. Keep its own image identities and experimental assignments. This article’s filenames and numerical results refer only to the eight supplied photographs.

The installed script examples/walkthrough_14_days.R follows the same numbered procedure. Precomputed tables in extdata/14_day_example are documented in their PROVENANCE.md. The steps above analyze the original photographs.

Comparison captions are exported separately as treatment_comparison_caption.txt. Use caption="brief" or caption="full" in plot_treatment_comparison() to include a caption beneath the figure.

Using your own photographs without completing experimental labels

The bundled example has matching labels already supplied, so it produces matched coefficients. Your own photographs can be measured even when treatment, set and optional annotations are blank. Prepare a new project, then run it directly. Unlabeled rows use the standard fungal layout; use default_centre="none" for control-only photographs. Mixed collections require the correct per-image centre_model or known control labels so the appropriate layout is used.

Finalization with an incomplete matching design saves measurements with comparison_status="awaiting_design" and an explanation in comparison_status.csv. It does not generate an unsupported comparison figure. Complete the actual treatment and own-control set assignments only when you need comparisons, then finish again. Do not remeasure unless the image layout changes. No optional strain, medium, date, notes or approval flags need filling merely to measure images. Inspect QC before interpretation. All displayed table examples use base R and knitr::kable(); no hidden tutorial helper needs to be copied. You can also use head(result$colonies) to inspect measurements immediately.

Explore measurements by set, without treatment names

You can inspect measurements using only set_id and the image identities. You do not need to name the treatments, fill optional annotations or change approval flags. These commands work after run_experiment(); they do not require a completed coefficient calculation. For your own ZIP, first run the standalone workflow below, then use its result14 and project14 objects here.

First, count the photographs recorded in each set:

metadata14 <- read.csv(project14$metadata_file, check.names=FALSE)
knitr::kable(as.data.frame(table(metadata14$set_id)),
             col.names=c("Set", "Photographs"), row.names=FALSE)

Set	Photographs
A60_2_14_days	4
B20_2_14_days	4

Next, average the finite colony measurements within each independent plate, then summarize those plate means by set. Each plate receives equal weight; colonies and spots are subsamples rather than independent plate replicates. Inspect the QC images and resolve failed photographs before interpretation. Keep different days, strains, media and other experimental contexts separate; subset the colony table to a single context before running this summary if those contexts occur within the same set.

colony_table14 <- result14$colonies
usable14 <- colony_table14[
  is.finite(colony_table14$MI_mean) &
    !is.na(colony_table14$set_id) & nzchar(trimws(colony_table14$set_id)),
  c("set_id", "plate_id", "MI_mean")
]
plate_means14 <- aggregate(MI_mean ~ set_id + plate_id,
                           data=usable14, FUN=mean)
knitr::kable(plate_means14, digits=3, row.names=FALSE)

set_id	plate_id	MI_mean
A60_2_14_days	2026_10_02_99_38	71.621
A60_2_14_days	2026_10_02_99_39	71.938
A60_2_14_days	2026_10_02_99_40	63.428
A60_2_14_days	2026_10_02_99_41	42.374
B20_2_14_days	2026_10_02_99_42	65.057
B20_2_14_days	2026_10_02_99_43	68.045
B20_2_14_days	2026_10_02_99_44	69.287
B20_2_14_days	2026_10_02_99_45	51.813

set_means14 <- aggregate(MI_mean ~ set_id, data=plate_means14, FUN=mean)
plate_counts14 <- aggregate(plate_id ~ set_id, data=plate_means14, FUN=length)
names(plate_counts14)[2] <- "n_measured_plates"
set_summary14 <- merge(set_means14, plate_counts14, by="set_id")
names(set_summary14)[names(set_summary14)=="MI_mean"] <- "mean_plate_MI"
knitr::kable(set_summary14, digits=3, row.names=FALSE)

set_id	mean_plate_MI	n_measured_plates
A60_2_14_days	62.34	4
B20_2_14_days	63.55	4

You can plot those plate means without assigning treatment names:

# Keep individual plates visible; use the final figure's colours and typography.
set_order14 <- sort(unique(as.character(plate_means14$set_id)))
set_palette14 <- stats::setNames(
  rep(c("#0072B2", "#D55E00", "#009E73", "#CC79A7", "#E69F00", "#56B4E9"),
      length.out=length(set_order14)), set_order14)
if (length(set_order14)>6L)
  set_palette14 <- stats::setNames(
    grDevices::hcl.colors(length(set_order14), "Dark 3"), set_order14)

points14 <- plate_means14[order(plate_means14$set_id, plate_means14$plate_id), ]
points14$row <- match(points14$set_id, set_order14)
for (set14 in set_order14) {
  ix14 <- which(points14$set_id==set14)
  points14$row[ix14] <- points14$row[ix14] +
    if (length(ix14)>1L) seq(0, 0.18, length.out=length(ix14)) else 0.08
}
set_sd14 <- aggregate(MI_mean ~ set_id, data=plate_means14, FUN=stats::sd)
names(set_sd14)[2] <- "sd_plate_MI"
summary_plot14 <- merge(set_summary14, set_sd14, by="set_id")
summary_plot14$row <- match(summary_plot14$set_id, set_order14)-0.22
intervals14 <- summary_plot14[is.finite(summary_plot14$sd_plate_MI), ]
labels14 <- paste0(set_order14, "\n",
  summary_plot14$n_measured_plates[match(set_order14, summary_plot14$set_id)],
  " measured plates")

set_plot14 <- ggplot2::ggplot(points14, ggplot2::aes(x=MI_mean, y=row)) +
  ggplot2::geom_segment(data=intervals14, ggplot2::aes(
    x=mean_plate_MI-sd_plate_MI, xend=mean_plate_MI+sd_plate_MI,
    y=row, yend=row, colour=set_id), inherit.aes=FALSE, linewidth=0.65) +
  ggplot2::geom_point(ggplot2::aes(colour=set_id, fill=set_id, shape="Plate"),
                      size=2.6, stroke=0.8) +
  ggplot2::geom_point(data=summary_plot14, ggplot2::aes(
    x=mean_plate_MI, y=row, colour=set_id, fill=set_id, shape="Set mean"),
    inherit.aes=FALSE, size=2.7) +
  ggplot2::scale_colour_manual(values=set_palette14, guide="none") +
  ggplot2::scale_fill_manual(values=set_palette14, guide="none") +
  ggplot2::scale_shape_manual(values=c("Plate"=21, "Set mean"=18)) +
  ggplot2::scale_y_continuous(breaks=seq_along(set_order14), labels=labels14,
    limits=c(0.5, length(set_order14)+0.5)) +
  ggplot2::labs(x="Plate melanization index", y=NULL, shape=NULL) +
  ggplot2::theme_classic(base_size=9, base_family="sans") +
  ggplot2::theme(
    axis.line.y=ggplot2::element_blank(), axis.ticks.y=ggplot2::element_blank(),
    axis.text=ggplot2::element_text(colour="#263238"),
    axis.text.y=ggplot2::element_text(size=8),
    panel.grid.major.y=ggplot2::element_line(colour="#E8ECEE", linewidth=0.3),
    legend.position="top", legend.justification="left",
    plot.margin=ggplot2::margin(10, 12, 10, 10))
print(set_plot14)

Circles show individual plates; the lower diamond shows the set mean. Bars show one between-plate SD, not confidence intervals, and are unavailable for a single plate. Colours identify sets, not treatment or control roles. This optional descriptive plot complements the two-panel matched-control figure in step 7; it does not replace it or infer treatment effects, coefficients or p-values. There is no caption embedded beneath either figure by default.

The plot displays in R and in this article. To save this optional figure, run:

for (ext14 in c("pdf", "svg", "png")) {
  ggplot2::ggsave(file.path(project14$project_dir,
                          paste0("set_measurements.", ext14)),
                 plot=set_plot14, width=7.2, height=3.8, dpi=600, bg="white",
                 device=if (ext14=="svg") grDevices::svg else
                   if (ext14=="pdf") grDevices::cairo_pdf else "png")
}

These are descriptive summaries of all measured plates in each set. If a set contains both control and confrontation plates, this summary includes both; it does not estimate the response to the interacting fungus. Blank set IDs are omitted, and plates without finite colony measurements are not counted. The count of successfully measured plates can therefore differ from the number of photographs in the metadata. Means based on incomplete colony detection need inspection as well.

set_id groups photographs but cannot identify which member is the control. For matched melanization coefficients, record the actual control in each set as treatment="control" and identify the other plates with a treatment code. A short code such as A60_2 is sufficient; a descriptive organism name is not required. Confirm the appropriate per-image layout, including centre_model="none" for control plates. If the layout changes, remeasure; otherwise save the assignments and run finish_experiment(project14) again. Without those assignments, awaiting_design is expected: measurements are available, but coefficient and treatment-comparison tables remain empty. Do not call melanization_coefficient() until the matching design is complete.

Standalone alternative: analyze your own 14-day ZIP

Use this complete alternative for your own archive, rather than mixing it with the eight-photo example above. The ZIP is the input file; the project directory is a separate new output folder. It must not be inside the ZIP filename. Do not rerun preparation over an existing project; reopen its project.rds with load_experiment() instead, or choose a new output folder.

library(FunMel)
photo_zip14 <- path.expand("~/Desktop/photos_14_days.zip")
project_dir14 <- path.expand("~/Desktop/FunMel_14_day_project")

# Reopen a prepared project; otherwise create it once from the ZIP.
if (file.exists(file.path(project_dir14, "project.rds"))) {
  project14 <- load_experiment(project_dir14)
} else {
  project14 <- prepare_experiment(
    input = photo_zip14,
    project_dir = project_dir14
  )
}

# Read existing measurements, or measure if this project has none yet.
if (file.exists(project14$result_file) || file.exists(project14$reviewed_file)) {
  result14 <- read_experiment(project14)
} else {
  # Replace 90 with the verified outer diameter of your dishes (mm).
  result14 <- run_experiment(
    project14, dish_diameter_mm=90,
    n_spots=5, spot_radius_mm=0.4
  )
}
result14$failed
head(result14$colonies)
head(result14$spots)
project14$qc_dir # Inspect all saved QC images before interpreting results.

# After resolving any failed photographs:
reviewed14 <- finish_experiment(project14)
reviewed14$comparison_status
comparison14 <- plot_treatment_comparison(reviewed14)
if (!is.null(comparison14)) print(comparison14)
project14$project_dir # Saved tables, analyses and any eligible comparison figures.

The two paths have different roles: photo_zip14 is the existing photograph ZIP, and project_dir14 is the output folder. "~" means your home folder; path.expand() converts it to a full path. file.path() joins folder names; it does not extract a ZIP or turn a ZIP file into a directory.

The block above can be rerun: it reopens project.rds when present and reads saved measurements when available. To import a different ZIP, choose a new output folder; reopening an existing project keeps that project’s photographs and metadata. To change the segmentation layout, dish scale or spot settings, call run_experiment() explicitly with the intended settings to replace the saved image measurements, then inspect the new QC and finalize again.

If the output folder is nonempty but has no project.rds, keep its contents and choose another new output folder; do not delete results to bypass the check. A failed assignment such as project14 <- prepare_experiment(...) does not clear or update an older project14 object. If preparation reports an existing project, reopen its actual folder:

project14 <- load_experiment(path.expand("~/Desktop/FunMel_14_day_project"))
project14$project_dir

Replace the input ZIP path with the actual location on your computer. A file chooser is an alternative: photo_zip14 <- file.choose(). The example paths above use the macOS Desktop; on other systems supply the correct full paths or use the chooser and a new output folder under path.expand("~"). The project14 object from this block refers to your own archive, not the bundled eight-photo example. Its photographs can have any supported filenames; do not reuse the bundled example’s fixed QC-image names or expected coefficients.

Treatment, set and optional annotations can stay blank for measurement. The fallback is a fungal confrontation layout. For unlabeled control-only images, add default_centre="none" to run_experiment(). Mixed collections require correct per-image centre_model values or verified control labels/folders. No manual approval flags or separate validation call are needed. Missing matching assignments produce measurement-only outputs with awaiting_design; add the real own-control assignments only when comparisons are needed. All metadata columns are preserved. Do not interpret a successful run as automatic biological validation; check masks, layouts, calibration and warnings.

When coefficient tables are empty

Colony melanization index (MI) and the matched melanization coefficient (MC) are different outputs. MI is stored in colonies.csv (and individual spot MI in spots.csv). MC divides treatment plate means by their own set’s control mean. A coefficient CSV with headers but no rows means no comparison was computed; inspect its recorded reason rather than treating it as zero.

knitr::kable(result14$run_summary, row.names=FALSE)

File	What to check
comparison_status.csv	Why comparisons are withheld or whether finalization is complete
metadata_issues.csv	Actual plate IDs, missing/inconsistent design fields and next steps
measurement_issues.csv	Plate and colony IDs with missing MI or incomplete requested spots
layout_assignments.csv	The layout actually used, and whether it came from metadata or a default
run_summary.csv	Image failures, finite target MI counts, incomplete spots and comparison status

An empty issue table means that no issue of that type was found by those checks; it does not establish biological validation. Diagnostic reports are separate from metadata: they do not overwrite custom annotations or approve any plate.

Understand the status before continuing

• analysis_status="complete" means all photographs were processed without a whole-image failure. An expected colony can still be undetected. For example, 20 processed plates can contain only 79 finite MI values among 80 target rows.
• comparison_status="awaiting_design" means required matching assignments are missing or inconsistent. Treatment names may be simple codes; real controls must be identified within each set. Other annotations may remain blank.
• comparison_status="awaiting_measurement_review" means a target colony is undetected or has nonfinite MI. Check the named QC image and correct its input or layout before reanalysis. The package saves measurements and withholds comparisons rather than inventing or silently omitting the missing value.
• Incomplete spots are reported separately. A colony with one of five requested spots can still have a valid whole-colony MI, but its requested spot summary is unavailable. Inspect masks and scale; use consistent feasible spot settings if remeasuring. Do not substitute a single achieved spot for five requested spots.

plot_treatment_comparison() explains the recorded reason and returns no comparison plot when the design or target measurements are unavailable. The set-only plot above remains descriptive and requires inspection of incomplete plates; it does not solve missing control assignments.

Assign controls from experimental records

The bundled eight photographs already have their design supplied. Your own ZIP may contain different photographs or additional sets: folder names such as A33_1_14_days identify a group but do not identify its control. Do not apply the bundled control assignments to unrelated images or guess that the last image in a folder is the control.

For a design with one experimental treatment per set, users who want to use set names as treatment codes can select their actual controls explicitly:

project <- load_experiment() # Open the existing project.rds.
metadata <- read.csv(project$metadata_file, check.names=FALSE,
                     stringsAsFactors=FALSE)
print(metadata[c("plate_id", "set_id", "treatment")], row.names=FALSE)

# Select only independently verified control photographs; one or more per set.
control_ids <- utils::select.list(metadata$plate_id, multiple=TRUE,
                                 title="Select the actual control plate IDs")
if (!length(control_ids)) stop("No controls selected; metadata was not changed.")

# Appropriate only when each set represents one experimental treatment.
metadata$treatment <- as.character(metadata$set_id)
metadata$treatment[metadata$plate_id %in% control_ids] <- "control"
metadata$centre_model <- ifelse(metadata$plate_id %in% control_ids,
                                "none", "fungus")
validate_metadata(metadata) # Check actual matching before saving these assignments.
write.csv(metadata, project$metadata_file, row.names=FALSE, na="")

This optional selection is an experimental assignment, not an approval flag. The code deliberately uses set names as treatment codes. If a set contains several different experimental treatments, supply their distinct true labels instead. All other metadata columns are retained. Cancel selection if the control identities are unknown; measurements can still be explored.

Decide whether image reanalysis is needed

If you only correct treatment/set labels and the saved image layouts were already correct, run finish_experiment(project) again; it reuses measurements.

If actual controls were processed using fungus, correcting those controls to none changes segmentation. Save the corrected metadata and rerun image analysis, inspect the new QC images, then finalize:

# Replace 90 with the verified dish diameter; keep the intended sampling settings.
result <- run_experiment(project, dish_diameter_mm=90, max_dim=1200,
                         n_spots=5, spot_radius_mm=0.4)
result$failed
result$run_summary
result$measurement_issues
project$qc_dir
# Resolve genuine failed/missing target measurements before interpreting comparisons.
reviewed <- finish_experiment(project)
comparison <- plot_treatment_comparison(reviewed)
if (!is.null(comparison)) print(comparison)

When a saved layout conflicts with a newly assigned control, the message names the affected plate IDs and explains that reanalysis is required. No missing measurement is filled from another colony or control, and no control is borrowed from a different set. Inspect the original photograph/QC when a colony cannot be resolved; software cannot establish a measurement for an absent or obscured target.