Technical documentation of the Shiny application
appDatos is a Shiny application designed for the interactive exploration of environmental and food safety laboratory data. It allows the user to:
library(shinyWidgets) # Enhanced widgets (pickerInput, actionBttn)
library(bslib) # Modern theme and layout (page_navbar, card, sidebar)
library(tidyverse) # Data manipulation and plotting (ggplot2, dplyr, stringr)
library(plotly) # Interactive charts (ggplotly)
library(shinycssloaders) # Loading spinner while plots render
library(DT) # Interactive tables
library(feather) # Fast file reading (loaded but not actively used)Data are stored as .rds files in the same directory as app.R. There are three file types per matrix:
| Type | Name pattern | Contents |
|---|---|---|
| Analytical data | <key>.rds |
Table with one column per analyte + Rótulo, Entidad, Fracción |
| Correlations | correl_<key>.rds |
Spearman correlation matrix between analytes |
| Analyte names | analisis_<key>.rds |
Character vector of available analyte names for the matrix |
matrices <- c(
"AGUA PROCESO", "AGUA SUPERFICIAL", "AGUA EFLUENTE", "AGUA SUMINISTRO",
"AGUA SUBTE", "SUELO REMEDIACIÓN", "JCONT LIMÓN", "JSIM LIMÓN",
"ACEITE LIMÓN", "DESH TE", "JCONT NARANJA", "JSIM NARANJA",
"JCONC MANZANA", "FF PERA", "FF MANZANA", "FF DURAZNO", "FF LIMÓN",
"BAYAS UVA", "HORTALIZA TOMATE", "VINO TINTO", "VINO BLANCO",
"VINO ROSADO", "PSIM DURAZNO", "PCON DURAZNO", "PSIM DAMASCO",
"PCON DAMASCO", "PSIM MANZANA", "PCON MANZANA", "CEREAL MAÍZ",
"TRUCHA MÚSCULO"
)The .rds files consumed by the application are produced by two functions defined in carga.R: carga_datos() for the analytical data tables and correlaciones() for the Spearman correlation matrices. Both share a common data preparation pipeline applied to a raw CSV exported from the LIMS.
The CSV is read with data.table::fread() for speed. Only the relevant columns are kept, and Resultado convertido is renamed to Resultado_conv.
datos <- fread(file = archivo) %>%
select(
Fracción, `Tipo de producto`, Matriz, Rótulo, Entidad, Análisis,
Resultado, `Modificador de resultado`, `Límite de detección`,
`Límite de cuantificación`, `Resultado convertido`, `Unidad inicial`
) %>%
rename(Resultado_conv = `Resultado convertido`) %>%
mutate(Resultado_conv = as.numeric(Resultado_conv))Results flagged as below the detection or quantification limit are replaced by the corresponding limit value:
datos <- datos %>%
mutate(Resultado_conv = case_when(
`Modificador de resultado` == "nd" ~ as.numeric(`Límite de detección`),
`Modificador de resultado` == "<" ~ as.numeric(`Límite de cuantificación`),
.default = Resultado_conv
))Some rows carry results expressed with a power-of-10 multiplier encoded in Unidad inicial (e.g. "x1³" means ×10³). These rows are separated, rescaled, and rejoined with the rest:
dat1 <- datos %>% filter(str_detect(`Unidad inicial`, "x1"))
supin <- str_extract(dat1$`Unidad inicial`, "\\W") # extract the superscript character
dat1 <- dat1 %>%
mutate(y = as.numeric(as.factor(supin))) %>% # encode as integer 1–6
mutate(Resultado = case_when(
y == 1 ~ Resultado * 1e3,
y == 2 ~ Resultado * 1e4,
y == 3 ~ Resultado * 1e5,
y == 4 ~ Resultado * 1e6,
y == 5 ~ Resultado * 1e7,
y == 6 ~ Resultado * 1e8
))
datos <- datos %>% filter(!str_detect(`Unidad inicial`, "x1"))
datos <- full_join(dat1, datos)Resultado_conv is the primary numeric value. When it is zero (i.e. originally missing), it falls back to Resultado. Zeros are then converted back to NA to mark missing data, and rows still missing a result are dropped:
datos <- datos %>%
select(Fracción, `Tipo de producto`, Matriz, Rótulo, Entidad,
Análisis, Resultado, Resultado_conv, `Unidad inicial`) %>%
replace(is.na(.), 0) %>%
mutate(Resultado_conv = ifelse(Resultado_conv == 0, Resultado, Resultado_conv)) %>%
replace(. == 0, NA) %>%
select(Fracción, `Tipo de producto`, Matriz, Rótulo, Entidad,
Análisis, Resultado_conv, `Unidad inicial`) %>%
drop_na()Three units are converted to a common scale before pivoting:
| Original unit | Multiplied by | Resulting unit |
|---|---|---|
| g/l | ×1 000 | mg/l |
| mS/cm | ×1 000 | µS/cm |
| g/kg | ×1 000 | mg/kg |
datos <- datos %>%
mutate(Resultado_conv = case_when(
`Unidad inicial` == "g/l" ~ Resultado_conv * 1000,
`Unidad inicial` == "mS/cm" ~ Resultado_conv * 1000,
`Unidad inicial` == "g/kg" ~ Resultado_conv * 1000,
.default = Resultado_conv
))Each unique value of Análisis becomes a column; the cell values are the consolidated Resultado_conv:
datos <- pivot_wider(datos, names_from = Análisis, values_from = Resultado_conv)The last digit of the Fracción code is standardised to "1" to group sub-fractions. Within each group, numeric columns are summed and Rótulo/Entidad are taken from the first row:
datos <- datos %>%
mutate(Fracción = str_replace_all(Fracción, "\\d$", "1")) %>%
replace(is.na(.), 0) %>%
group_by(Fracción) %>%
summarise(
Rótulo = first(Rótulo),
Entidad = first(Entidad),
across(where(is.numeric), sum)
) %>%
ungroup()Entities are sorted alphabetically and stored as an ordered factor. Zeros introduced during aggregation are converted back to NA:
r <- sort(unique(datos$Entidad))
datos <- datos %>%
mutate(Entidad = factor(Entidad, levels = r)) %>%
replace(. == 0, NA)carga_datos(): saving the analytical tableAfter the common pipeline, the wide-format tibble is saved directly as an .rds file ready for the application:
saveRDS(datos, guardado)correlaciones(): computing and saving the Spearman matrixAfter the same pipeline, correlaciones() additionally:
Fracción, Rótulo, and Entidad identifier columns to keep only the analyte columns.NA observation (required for a valid correlation estimate).cor(..., use = "pairwise.complete.obs").Analisis name column, and saves as .rds.data <- datos %>% select(-(1:3))
# Keep only analytes with more than one observation
x <- sapply(1:ncol(data), function(i) data %>% filter(!is.na(data[, i])) %>% nrow())
data1 <- data[, which(x > 1)]
matriz <- cor(data1, method = "spearman", use = "pairwise.complete.obs")
correl <- as_tibble(matriz) %>% round(2)
correl <- correl %>% mutate(Analisis = names(correl))
write_rds(correl, archivo_output)Note:
correlaciones()also writes ananalisis_*.rdsfile containing the vector of analyte names. Incarga.Rthis filename is hard-coded ("analisis_trucha_musc.rds"); in practice it would need to be parameterised the same wayarchivo_outputis.
The UI is built with bslib::page_navbar(), which produces a top navigation bar with tabs and a collapsible sidebar.
page_navbar
├── nav_panel("Dos variables") ← Tab: scatter plot + correlation table
├── nav_panel("Una variable") ← Tab: histogram
├── nav_item: selectizeInput(matrices) ← Matrix selector (embedded in navbar)
├── nav_item: uiOutput(selec_analisis_x) ← X-axis selector (dynamic)
├── nav_item: conditionalPanel ← Y-axis selector (only in "Dos variables")
└── sidebar
├── ("Una variable" panel) textInput(rot1) + actionBttn(buscar1)
├── ("Dos variables" panel) textInput(rot) + actionBttn(buscar)
├── checkboxInput(log) ← Apply log10() to BOTH axes in the data values
├── checkboxInput(logy) ← Log10 axis scale on Y (axis transformation)
├── checkboxInput(logx) ← Log10 axis scale on X (axis transformation)
├── numericInput(vert, vert2) ← Vertical reference lines at x
├── ("Dos variables") numericInput(hor, hor2) ← Horizontal reference lines at y
├── checkboxInput(ent) ← Color points/bars by entity
├── uiOutput(selec_entidades) ← Multi-select entity picker (dynamic)
└── ("Dos variables") numericInput(pendiente, ordenada, ...) ← Manual regression linesconditionalPanel() is used to show or hide controls based on the active tab (input.nav):
Three UI outputs are rendered dynamically by the server because they depend on the loaded data:
| Output | Widget produced | Depends on |
|---|---|---|
selec_analisis_x |
selectizeInput (X axis) |
analisis() — analyte names for the matrix |
selec_analisis_y |
selectizeInput (Y axis) |
analisis() — same as above |
selec_entidades |
pickerInput (multi-select) |
datos_full()$Entidad — unique entities in the data |
input$matrices
│
├──► datos_full() ─────────────────────────────────────────────────┐
│ (analytical data for the selected matrix) │
│ │
├──► analisis() ──► selec_analisis_x / selec_analisis_y (UI) │
│ (analyte names) │
│ │
└──► correl() ──► tablacorr (correlation table) │
(Spearman correlation matrix) │
│
input$buscar ──► rotulos() ────────────────────► subsetted() ──► scatterplot
(label filter) │
│ also filters by entity
input$enti ────────────────────────────────────────────┘
input$buscar1 ──► rotulos1() ───────────────────► subsetted1() ──► scatterplot2
(label filter for histogram)datos_full(), analisis(), correl()All three are eventReactive(input$matrices, {...}): they recompute only when the user changes the matrix. They use switch() to map the matrix name to the corresponding .rds file.
datos_full <- eventReactive(input$matrices, {
req(input$matrices)
switch(input$matrices,
"AGUA PROCESO" = read_rds("agupr.rds"),
"AGUA SUPERFICIAL" = read_rds("super_agua.rds"),
# ... (30 matrices)
)
})Note: The
analisis_*andcorrel_*objects are loaded once at session startup (outside the server function). Theanalisis()andcorrel()reactives simply select the already-loaded object from memory viaswitch(), without reading files on every matrix change.
rotulos() and rotulos1()Both are eventReactive triggered by the “Visualizar” button (input$buscar or input$buscar1). This is an intentional design choice: the plot does not update as the user types, only when the button is clicked.
rotulos <- eventReactive(input$buscar, {
if (input$rot != "") {
datos_full() %>%
filter(str_detect(Rótulo, regex(input$rot, ignore_case = TRUE))) %>%
pull(Rótulo)
} else {
datos_full() %>% pull(Rótulo) # No filter: returns all labels
}
})The search uses str_detect() with regex(..., ignore_case = TRUE), so it accepts regular expressions and is case-insensitive.
subsetted() and subsetted1()These combine the entity filter (multi-select picker) and the label filter:
subsetted <- reactive({
req(input$enti)
datos_full() %>%
filter(Entidad %in% input$enti & Rótulo %in% rotulos())
})subsetted() feeds the scatter plot; subsetted1() feeds the histogram.
output$scatterplot)Built with ggplot2 and made interactive with ggplotly(). The plot only updates when input$buscar is pressed, because it depends on rotulos().
ggplot(subsetted(), aes(.data[[input$x]], .data[[input$y]],
text = paste0("Fracción: ", Fracción, "</br>",
"Rótulo: ", Rótulo))) +
geom_point() +
geom_vline(xintercept = input$vert, color = "red") +
geom_vline(xintercept = input$vert2, color = "red") +
geom_hline(yintercept = input$hor, color = "red") +
geom_hline(yintercept = input$hor2, color = "red") +
geom_abline(slope = input$pendiente, intercept = input$ordenada, color = "green") +
geom_abline(slope = input$pendiente2, intercept = input$ordenada2, color = "green")The server covers all combinations of three boolean inputs via explicit if/else if branches:
input$ent |
input$logx |
input$logy |
input$log |
Effect |
|---|---|---|---|---|
| FALSE | FALSE | FALSE | FALSE | Standard plot, no entity coloring |
| FALSE | TRUE | FALSE | FALSE | X axis on log10 scale |
| FALSE | TRUE | TRUE | FALSE | Both axes on log10 scale |
| FALSE | FALSE | TRUE | FALSE | Y axis on log10 scale |
| TRUE | FALSE | FALSE | FALSE | Points colored by Entidad |
| TRUE | TRUE | FALSE | FALSE | Entity color + X axis log10 |
| TRUE | FALSE | TRUE | FALSE | Entity color + Y axis log10 |
| TRUE | TRUE | TRUE | FALSE | Entity color + both axes log10 |
| TRUE/FALSE | — | — | TRUE | log10() applied to the data values (not the axis scale) |
Key distinction:
input$logx/input$logyusescale_x/y_continuous(trans = "log10"), which transforms the axis scale while keeping the original values in the tooltip. In contrast,input$logapplieslog10()directly insideaes(), transforming the data before plotting — so axis labels show the log-transformed values.
output$scatterplot2)Uses only the X axis (input$x) to show the distribution of a single analyte. Also made interactive via ggplotly().
ggplot(subsetted1(), aes(.data[[input$x]],
text = paste0("Fracción: ", Fracción, "</br>",
"Rótulo: ", Rótulo))) +
geom_histogram() +
geom_vline(xintercept = input$vert, color = "red") +
geom_vline(xintercept = input$vert2, color = "red") +
ylab("N° de muestras")Variants: combination of input$ent (entity coloring via fill = Entidad) and input$logx (log10 scale on X axis), yielding four branches.
output$tablacorr)Displays the analytes most correlated with the selected X-axis analyte, filtered to |r| > 0.5 and sorted in descending order:
output$tablacorr <- renderDT({
req(input$x, correl())
correl() %>%
filter(.data[[input$x]] > 0.5 | .data[[input$x]] < -0.5) %>%
select(Analisis, all_of(input$x)) %>%
arrange(desc(.data[[input$x]])) %>%
datatable()
})The table header is dynamic (output$encabezado) and displays the name of the currently selected X analyte.
User selects a matrix
│
▼
datos_full() + analisis() + correl() ← loaded / selected from memory
│
▼
X/Y axis selectors and entity picker update (renderUI)
│
User chooses analytes, types a label filter, selects entities
│
▼
Clicks "Visualizar"
│
▼
rotulos() / rotulos1() ← label filter applied
│
▼
subsetted() / subsetted1() ← data filtered by entity and label
│
├──► scatterplot ("Dos variables" tab) + tablacorr
└──► scatterplot2 ("Una variable" tab / histogram)analisis_* vectors and correl_* matrices are loaded once at startup to minimize latency when switching matrices.if/else if branches in the scatter plot renderer could be simplified by building the base plot once and appending layers conditionally.switch() blocks (data, analisis, correl) all share the same matrix list and could be consolidated into a helper function or a named lookup list.feather package is loaded but never used and can be removed.