BRIDGE

BRIDGE is a framework for deconvolving bulk expression data into molecular subtype fractions and using these profiles to predict therapy response. All data and code to reproduce the paper can be found at https://hpc.nih.gov/~Lab_ruppin/BRIDGE_main_scripts.zip.

You can find the full research paper here

---

Installation

You can install BRIDGE directly from GitHub:

# Install devtools if not already installed
install.packages("devtools")

# Install BRIDGE
devtools::install_github("cantorethomas/BRIDGE")

How to run

1. Derive subtype abundances BRIDGEdeconv()

Please Note: BRIDGEdeconv() can be run using

• reference='PAM50'
• reference='TNBC'

library(BRIDGE)

# Synthetic expression matrix (genes x samples)

set.seed(3)
gsel <- sample(rownames(reference_BRIDGE_PAM50), 600)
expr <- matrix(round(runif(600 * 3), 3) * 1000, nrow = 600,
               dimnames = list(gsel, paste0("Sample", 1:3)))

# Run BRIDGE deconvolution
res <- BRIDGEdeconv(expr_matrix = expr, reference = "PAM50")

The function returns a numeric matrix with dimensions (samples × subtypes).

• Rows: individual samples from the input expression matrix
• Columns: molecular subtypes defined by the chosen reference (e.g., PAM50)
• Values: estimated relative abundance of each subtype within each sample

Example:

Sample	BASA	LUMA	LUMB	HER2
P1	0.70	0.10	0.05	0.10
P2	0.15	0.40	0.30	0.10
P3	0.05	0.50	0.35	0.05

---

2. Deconvolution and response prediction BRIDGEpredict()

⚠️ Research use only. BRIDGE is intended strictly for research purposes and has not been validated for clinical decision-making.

BRIDGEpredict() requires two arguments: subtype and therapy. The following combinations are currently supported:

subtype	therapy	Model type
ERpos_HER2neg	CHEMO	Main
ERpos_HER2neg	IMMUNO	Exploratory
ERpos	ENDO	Exploratory
HER2pos	ANTI_HER2	Main
TNBC	CHEMO	Main
TNBC	IMMUNO	Exploratory

Main models have been trained and validated on multiple cohorts. Exploratory models are based on limited data and should be interpreted with caution.

# Run BRIDGE predictive model for a given subtype and therapy
res_pred <- BRIDGEpredict(expr_matrix = expr,
                          subtype     = "ERpos_HER2neg",
                          therapy     = "CHEMO")

# Access results
fractions <- res_pred$fractions
scores    <- res_pred$BRIDGE_SCORE

head(fractions)
head(scores)

The function returns a list with two elements:

1. fractions
   Same output as BRIDGEdeconv() (samples × subtypes matrix). Useful for exploring intra-tumoral heterogeneity independently of the predictive model.

2. BRIDGE_SCORE
   A data frame with three columns:

   Column	Description
   SCORE	Continuous response score (0–1)
   CLASS	Predicted class: "high" or "low" response
   SUBCLASS	Quantile-based sub-class (q1–q5); NA if unavailable

   Example output:

   	SCORE	CLASS	SUBCLASS
   P1	0.82	high	q5
   P2	0.47	low	q3
   P3	0.65	high	q4

BRIDGE-Slide

BRIDGE-Slide enables direct application of BRIDGE from histopathology slides. The workflow first applies Path2Omics to infer cohort-level gene expression profiles from H&E slides, followed by BRIDGE to predict molecular subtype composition and therapy response.

For running Path2Omics please see: Path2Omics GitHub repository