FAQs
Frequently Asked Questions
General¶
What does Pheno-Ranker do?
Pheno-Ranker is an open-source toolkit developed for the semantic similarity analysis of phenotypic and clinical data. It natively supports GA4GH standards, such as Phenopackets v2 and Beacon v2, using as input their JSON/YAML data exchange formats. Beyond these specific standards, Pheno-Ranker is designed to be highly versatile, capable of handling any data serialized into JSON, YAML, and CSV formats, extending its utility beyond the health data domain. Pheno-Ranker transforms hierarchical data into binary digit strings, enabling efficient similarity matching both within cohorts and between individual patients and reference cohorts.
last change 2024-09-23 by Manuel Rueda ¶
Where can I find the Web App user interface?
You can find the Web App user interface at this address: https://pheno-ranker.cnag.eu.
There is a playground that you can use by entering the following credentials:
Username: pheno
Password: ranker
last change 2024-04-01 by Manuel Rueda ¶
Can I export term coverage and intermediate files?
Yes. It is possible to export a file indicating coverage for each term (i.e., 1D-keys) as well as all intermediate files using the flag --e.
On top of that, in patient mode, alignment files can be obtained by using --align.
last change 2023-10-13 by Manuel Rueda ¶
How can I exclude a given variable?
To exclude a specific variable, you can use one of the following methods:
- Utilize the
--include-termsor--exclude-termsoptions in the command-line interface (CLI). - Implement a regular expression (regex) in the configuration file using the
exclude_variables_regexparameter. - Assign a weight of zero to the variable in a weights file (indicated by the
--wflag). This approach offers the most detailed control over variable exclusion.
last change 2023-12-22 by Manuel Rueda ¶
Can I use pedigrees term in BFF?
A priori, you can, but the term pedigrees is excluded by default via configuration file. Pedigrees are often case-related, so the information is not relevant for comparison to other cases. Additionally, it contains deeply nested data structures. If you want to include it, please modify the default configuration file and use it with the --config <your-config-file> option.
last change 2024-09-27 by Manuel Rueda ¶
How does Pheno-Ranker treat empty and missing values?
Pheno-Ranker uses categorical variables to define the binary_digit_vector. Any key that contains an empty value such as null, {}, or [] is discarded. We also deliberately discard keys with missing values (namely NA and NaN). On the other hand, any other string such as Unknown or No is accepted as values and will be incorporated as a category. Thus, if you want missing values to be included as a category, we recommend replacing them with another string (e.g., Unknown).
Now, in the context of one-hot encoded data, missing values are typically represented by zeros, and that's exactly Pheno-Ranker's approach.
Let's consider the following example:
[
{
"id": "PR_1",
"Name": "Foo"
},
{
"id": "PR_2",
"Name": "Bar"
},
{
"id": "PR_3"
},
{
"id": "PR_4",
"Name": null
},
{
"id": "PR_5",
"Name": "NA"
}
]
We have only one variable (i.e., category) named Name. The first two individuals contain valid values, the third does not contain the key, the fourth has it as null, and the fifth has NA.
The coverage (see "Can I export term coverage and intermediate files?" above) for the terms will be the following:
In which we see that the variable Name only has coverage for 2 out of 5 individuals.
If we run a job with the default metric (Hamming distance) and --include-terms Name, the global hash will look like this:
The binary-digit-vector will look like this:
The resulting matrix.txt will look like this:
| PR_1 | PR_2 | PR_3 | PR_4 | PR_5 | |
|---|---|---|---|---|---|
| PR_1 | 0 | 2 | 1 | 1 | 1 |
| PR_2 | 2 | 0 | 1 | 1 | 1 |
| PR_3 | 1 | 1 | 0 | 0 | 0 |
| PR_4 | 1 | 1 | 0 | 0 | 0 |
| PR_5 | 1 | 1 | 0 | 0 | 0 |
In this context, the distance from PR_1 to PR_2 is 2 (indicating they differ in two positions), while the distance from PR_1 to PR_3, PR_4, and PR_5 is only 1. This discrepancy arises because an empty value contributes less to the distance calculation than a non-empty one. While this discrepancy may often be acceptable since the data will still fall into different clusters, an alternative solution to mitigate this issue is to exclude the variable Name entirely when running Pheno-Ranker. Another, less drastic approach is to employ the Jaccard metric (--similarity-metric-cohort jaccard), which is less sensitive to empty values. With the Jaccard index, the resulting matrix will be as follows:
| PR_1 | PR_2 | PR_3 | PR_4 | PR_5 | |
|---|---|---|---|---|---|
| PR_1 | 1 | 0.000000 | 0.000000 | 0.000000 | 0.000000 |
| PR_2 | 0.000000 | 1 | 0.000000 | 0.000000 | 0.000000 |
| PR_3 | 0.000000 | 0.000000 | 1 | 0.000000 | 0.000000 |
| PR_4 | 0.000000 | 0.000000 | 0.000000 | 1 | 0.000000 |
| PR_5 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | 1 |
When handling missing values within the context of Pheno-Ranker, it's essential to consider the specific characteristics of your dataset, the nature of missingness, and the potential impact on downstream analyses
last change 2024-08-04 by Manuel Rueda ¶
Pre-processing¶
How can I create a JSON file consisting of a subset of individuals?
You can use the tool jq:
# Let's assume you have an array of "id" values in a variable named ids
ids=( "157a:week_0_arm_1" "157a:week_2_arm_1" )
# Use jq to filter the array based on the "id" values
jq --argjson ids "$(printf '%s\n' "${ids[@]}" | jq -R -s -c 'split("\n")')" 'map(select(.id | IN($ids[])))' < individuals.json > subset.json
last change 2024-07-02 by Manuel Rueda ¶
I have noticed that in cohort mode, Pheno-Ranker takes as input an array of objects. Does it also support independent JSON files (one per patient)?
The simple answer is yes. However, what actually happens under the hood is that each independent file is treated as a cohort, and a prefix (defaulting to 'CX_') is added to the primary_key ID. This does not affect the results. Additionally, in colored MDS plots, each patient represented in separate files will be distinguished with a different color.
If you prefer to combine all independent JSON files into a single JSON array, consider using one of the following alternatives:
With the tool jq:
Alternatively, if you want to resort to Bash:
See Bash code:
last change 2024-04-16 by Manuel Rueda ¶
Do you account for BFF/PXF schema versions?
As of August 2024, we do not explicitly account for BFF/PXF schema versions. In some cases, BFF data may not include a version, requiring us to infer changes. However, most schema updates are downstream and do not impact term-level data. As a result, the overall effect of comparing data from different schema versions is typically minimal. We assume that users are aware of the versions they are working with and understand the implications of using data from different schema versions.
last change 2024-08-11 by Manuel Rueda ¶
Post-processing¶
How do I store Pheno-Ranker's data in a relational database?
First, export intermediate files using the following command:
This command will generate a set of intermediate files, and the one you'll need for queries is my_export_data.ref_binary_hash.json.
To convert the JSON data to CSV, you can use various methods, but we recommend using the jq tool:
jq -r 'to_entries[] | [.key, .value.binary_digit_string] | @csv' < my_export_data.ref_binary_hash.json | awk 'BEGIN {print "id,binary_digit_string"}{print}' > output.csv
The results are now stored at output.csv
Finally, store the data in your database as you will usually do.
last change 2023-10-13 by Manuel Rueda ¶
Can I Perform MDS with Jaccard Indices?
Yes, you can perform Multidimensional Scaling (MDS) using a matrix of Jaccard indices. To use MDS, which typically requires dissimilarity data, you'll need to convert your Jaccard similarity matrix into a dissimilarity matrix. This is done by subtracting the Jaccard similarity scores from 1, where the formula is:
$$ D = 1 - J $$ Where: \( D \) is the dissimilarity. \( J \) is the Jaccard index.
This conversion ensures that higher similarities translate into shorter distances for MDS, facilitating accurate low-dimensional representations of the data.
Example R code:
# Load the matrix of Jaccard similarities from a text file
data <- as.matrix(read.table("matrix.txt", header = TRUE, row.names = 1))
# Convert Jaccard similarity matrix to a dissimilarity (distance) matrix
dissimilarity_matrix <- 1 - data
# Perform classical Multidimensional Scaling (MDS) using the dissimilarity matrix
# 'eig=TRUE' allows the function to return eigenvalues
# 'k=2' sets the number of dimensions for the MDS output
fit <- cmdscale(dissimilarity_matrix, eig=TRUE, k=2)
# Additional analysis and plotting code here
...
last change 2024-04-15 by Manuel Rueda ¶
Can I convert a Hamming distance matrix to a similarity matrix?
First of all, if you are seeking a similarity metric, you might want to consider using jaccard as a metric. However, if you wish to convert a distance-based matrix to a similarity matrix, you can use the following formula:
$$ S = 1 - \frac{d}{n} $$ Where: \( S \) is the similarity. \( d \) is the Hamming distance. \( n \) is the number of characters compared in the Hamming distance.
Example R code:
# Load the matrix of Hamming distances from a text file
data <- as.matrix(read.table("matrix.txt", header = TRUE, row.names = 1))
# Set n (extracted with --export option)
n = 100
# Convert Hamming distance to a similarity matrix
similarity_matrix <- 1 - (data / n)
# Additional analysis and plotting code here
...
last change 2024-04-15 by Manuel Rueda ¶
Can I create network/graph plots from Pheno-Ranker output data?
Absolutely, you can—the possibilities are endless! 
Pheno-Ranker can generate graph data in JSON format which is compatible with the Cytoscape ecosystem. To create a graph, you can execute the following command:
If you like to get summary statistics for the graph use it in conjunction with --graph-stats, like this:
pheno-ranker -r individuals.json --cytoscape-json cytoscape_graph.json --graph-stats my_graph_stats.txt
The file my_graph_stats.txt will include summary statistics and the shortest path between all nodes. Be aware that this calculation may be time-consuming for large graphs.
Alternatively, you can use R for more graphical options. Here are some examples using the qgraph and igraph packages:
Reference cohort
See code
First, we run Pheno-Ranker in cohort mode using jaccard as a metric:
Now we plot the resulting matrix.txt file.
Coloring Nodes and Edges:¶
- Nodes: Colored based on the count of connections exceeding a specified threshold, using a gradient from red (fewer connections) to blue (more connections).
- Edges: Colored by weight, with blue for the strongest connections (weight > 0.90), green for strong connections (weight > 0.50), and red for weaker ones.
# Install the necessary packages if not already installed
if (!requireNamespace("qgraph", quietly = TRUE)) {
install.packages("qgraph")
}
library(qgraph)
data <- as.matrix(read.table("matrix.txt", header = TRUE, row.names = 1, check.names=FALSE))
# Start PNG device
png(filename = "qgraph.png", width = 1000, height = 1000,
units = "px", pointsize = 12, bg = "white", res = NA)
# Toggle for coloring the last node black
colorLastNodeBlack <- FALSE # Change to TRUE for black / FALSE to retain original color
# Function to determine color based on threshold values
getColorBasedOnThreshold <- function(value, thresholdHigh, thresholdMid, colorHigh, colorMid, colorLow) {
if (value > thresholdHigh) {
return(colorHigh)
} else if (value > thresholdMid) {
return(colorMid)
} else {
return(colorLow)
}
}
# Apply this function to each node and edge
node_thresholds <- apply(data, 1, function(x) sum(x > 0.9))
max_node_threshold <- max(node_thresholds)
min_node_threshold <- min(node_thresholds)
normalized_node_thresholds <- (node_thresholds - min_node_threshold) / (max_node_threshold - min_node_threshold)
# Color nodes based on normalized threshold
node_colors <- colorRampPalette(c("red", "green", "blue"))(length(unique(normalized_node_thresholds)))
node_colors <- node_colors[as.integer(cut(normalized_node_thresholds, breaks = length(node_colors), include.lowest = TRUE))]
# Conditionally color the last node black
if (colorLastNodeBlack) {
node_colors[length(node_colors)] <- "black" # Last node in black
}
# Edge colors with similar logic
edge_colors <- apply(data, c(1,2), function(x) getColorBasedOnThreshold(x, 0.90, 0.50, "blue", "green", "red"))
edge_colors <- matrix(edge_colors, nrow=nrow(data), ncol=ncol(data))
# Create and plot the graph
qgraph(data,
labels=colnames(data),
layout='spring',
label.font=2, # Bold labels
vsize=10, # Node size
threshold=0.50, # Edge visibility threshold
shape='circle',
color=node_colors, # Node colors
edge.color=edge_colors, # Edge colors
edge.width=1) # Edge width
# Close the device to save the PNG file
dev.off()
Reference cohort - Target patient
See code
Again, we run Pheno-Ranker in cohort mode, adding individuals.json and patient.json as if they were two cohorts, using jaccard as a metric:
pheno-ranker -r individuals.json patient.json --append-prefixes REF TAR --similarity-metric-cohort jaccard
Now we plot the resulting matrix.txt file, but this time the TAR_107:week_0_arm_1 (last element) node is colored black to be more visible.
Coloring Nodes and Edges:¶
- Nodes: Colored based on the count of connections exceeding a specified threshold, using a gradient from red (fewer connections) to blue (more connections).
- Edges: Colored by weight, with blue for the strongest connections (weight > 0.90), green for strong connections (weight > 0.50), and red for weaker ones.
# Install the necessary packages if not already installed
if (!requireNamespace("qgraph", quietly = TRUE)) {
install.packages("qgraph")
}
library(qgraph)
data <- as.matrix(read.table("matrix.txt", header = TRUE, row.names = 1, check.names=FALSE))
# Start PNG device
png(filename = "qgraph.png", width = 1000, height = 1000,
units = "px", pointsize = 12, bg = "white", res = NA)
# Toggle for coloring the last node black
colorLastNodeBlack <- TRUE # Change to TRUE for black / FALSE to retain original color
# Function to determine color based on threshold values
getColorBasedOnThreshold <- function(value, thresholdHigh, thresholdMid, colorHigh, colorMid, colorLow) {
if (value > thresholdHigh) {
return(colorHigh)
} else if (value > thresholdMid) {
return(colorMid)
} else {
return(colorLow)
}
}
# Apply this function to each node and edge
node_thresholds <- apply(data, 1, function(x) sum(x > 0.9))
max_node_threshold <- max(node_thresholds)
min_node_threshold <- min(node_thresholds)
normalized_node_thresholds <- (node_thresholds - min_node_threshold) / (max_node_threshold - min_node_threshold)
# Color nodes based on normalized threshold
node_colors <- colorRampPalette(c("red", "green", "blue"))(length(unique(normalized_node_thresholds)))
node_colors <- node_colors[as.integer(cut(normalized_node_thresholds, breaks = length(node_colors), include.lowest = TRUE))]
# Conditionally color the last node black
if (colorLastNodeBlack) {
node_colors[length(node_colors)] <- "black" # Last node in black
}
# Edge colors with similar logic
edge_colors <- apply(data, c(1,2), function(x) getColorBasedOnThreshold(x, 0.90, 0.50, "blue", "green", "red"))
edge_colors <- matrix(edge_colors, nrow=nrow(data), ncol=ncol(data))
# Create and plot the graph
qgraph(data,
labels=colnames(data),
layout='spring',
label.font=2, # Bold labels
vsize=10, # Node size
threshold=0.50, # Edge visibility threshold
shape='circle',
color=node_colors, # Node colors
edge.color=edge_colors, # Edge colors
edge.width=1) # Edge width
# Close the device to save the PNG file
dev.off()
Reference cohort - Shortest path between two individuals
See code
Imagine that Pheno-Ranker has created a matrix.txt with this content:
| PR_1 | PR_2 | PR_3 | PR_4 | PR_5 | |
|---|---|---|---|---|---|
| PR_1 | 0 | 9 | 4 | 4 | 6 |
| PR_2 | 9 | 0 | 2 | 4 | 6 |
| PR_3 | 4 | 2 | 0 | 4 | 6 |
| PR_4 | 4 | 4 | 4 | 0 | 2 |
| PR_5 | 6 | 6 | 6 | 2 | 0 |
You want to know the shortest path between PR_1 and PR_2 (Hamming distance = 9). Let's process it with R:
# Install the necessary packages if not already installed
if (!requireNamespace("igraph", quietly = TRUE)) {
install.packages("igraph")
}
library(igraph)
# Define start and end nodes
START_NODE <- "PR_1"
END_NODE <- "PR_2"
# Read the data matrix from a file, assuming it's a distance matrix
data <- as.matrix(read.table("matrix.txt", header = TRUE, row.names = 1, check.names = FALSE))
# Adjust weights of zero to a small positive value for calculation
data[data == 0] <- 1e-10
# Create an igraph graph from the adjusted distance matrix
g <- graph_from_adjacency_matrix(data, mode = "undirected", weighted = TRUE, diag = FALSE)
# Set node names
V(g)$label <- V(g)$name # assuming node names are already defined
# Find the shortest paths between START_NODE and END_NODE
shortest_path <- shortest_paths(g, from = START_NODE, to = END_NODE, mode = "out", output = "vpath")
# Extract the edges in the shortest path
edges_in_path <- E(g, path = unlist(shortest_path$vpath))
# Define colors for vertices and edges
vertex_colors <- ifelse(V(g)$name == START_NODE, "orange", ifelse(V(g)$name == END_NODE, "green", "grey"))
edge_colors <- ifelse(E(g) %in% edges_in_path, "black", "grey")
edge_widths <- ifelse(E(g) %in% edges_in_path, 3, 0.2)
# Prepare edge labels to display original values
edge_labels <- ifelse(E(g)$weight == 1e-10, 0, E(g)$weight)
# Start PNG device
png(filename = "igraph.png", width = 1000, height = 1000,
units = "px", pointsize = 12, bg = "white", res = NA)
# Plot the graph
plot(g, layout = layout_nicely(g),
edge.label.cex = 3, # Edge label size
edge.color = edge_colors,
edge.width = edge_widths,
edge.label = edge_labels, # Use adjusted labels for display
label.font=2, # Bold labels
label.distance = 1,
vertex.color = vertex_colors,
vertex.size = 40, # Increased node size
vertex.label.cex = 3.0, # Size of labels
vertex.label.color = "black", # Label color
vertex.label.font = 2, # Bold labels
vertex.label.family = "sans", # Font family
vertex.label.fontcolor = "black") # Font color
dev.off()
last change 2024-04-15 by Manuel Rueda ¶
Installation¶
From where can I download the software?
Should you opt for the command-line interface, we suggest obtaining the software from the CPAN distribution, which additionally includes the utility bff-pxf-simulator.