DualMarker

DualMarker: A multi-source fusion identification method for prognostic biomarkers of breast cancer based on dual-layer heterogeneous network.

Description

Getting Started

Introduction

DualMarker is a biomarker recognition algorithm.

Special dependencies

• MATLAB R2020a or later (R2021b+ recommended)
• Statistics and Machine Learning Toolbox (for standardization & classifier; you can also use TreeBagger)

---

Overview

The repository is organized as follows:

• DualMarkerMainCode.m

  Main pipeline of DualMarker :

  1. Build the heterogeneous network and compute RandNE embeddings;
  2. Derive prior scores from DisGeNET (gene–disease) and expression (t-test), normalize and fuse;
  3. Perform random walk / network propagation on the dual-layer network;
  4. Output final biomarker ranking in variable geneRANK.

• KFoldCrossValidation.m

  Repeated 5-fold cross-validation (default 50 repeats × 5 folds ) on Top-100/150/200 features selected from geneRANK. Trains a classifier (random forest by default) and reports AUC for each repeat as well as the final mean AUCs.

• Core utilities

  A_RWRplus.m, Network_Enhancement.m, getNormalizedMatrix_Heter.m,

  getRandNEemb_in.m / RandNE_*, Obtain_StatisticScore.m, TransitionFields.m, GS.m, …

• Data/

  One packaged .mat file with everything needed to run DualMarker.

---

Data

We provide a packaged file (example name):

Data/initavalue2.mat

This .mat file bundles all inputs needed to run DualMarker:

• Heterogeneous network adjacency: a single sparse block matrix encoding edges among genes, proteins, diseases, and GO terms (e.g., gene–gene, gene–protein, protein–protein, gene–disease, gene–GO). Used to build the dual-layer network and for propagation.
• Expression matrix & gene list: normalized gene expression values (samples × genes) with matching gene identifiers. Used to compute differential signals and for 5-fold cross-validation.
• DisGeNET prior: disease–gene relevance scores that act as prior seeds to guide the propagation and ranking.

Running DualMarker

1) Produce the biomarker ranking

addpath(genpath(pwd));                % add repo to MATLAB path
load('Data/DualMarker_demo.mat');     % load the packaged inputs

run('DualMarkerMainCode.m');          % produces variable "geneRANK"
head(geneRANK);                       % [gene, score] sorted descending

2) 5-fold CV (50 repeats) at Top-100/150/200

% Requires "geneRANK" from the step above
run('KFoldCrossValidation.m');

(Optional) Save the AUC lists

writematrix(GSE2034_aucScore_list_100, 'GSE2034_top100_auc.csv');
writematrix(GSE2034_aucScore_list_150, 'GSE2034_top150_auc.csv');
writematrix(GSE2034_aucScore_list_200,  'GSE2034_top200_auc.csv');

---

Evaluating DualMarker

• Top-K feature selection is handled inside KFoldCrossValidation.m via geneRANK(1:K).
• Classifier : by default we use a random forest (classRF_*). To use MATLAB’s native model:

  % Example replacement using TreeBagger
  model = TreeBagger(500, train_X', categorical(train_Y));
  [~, score] = predict(model, test_X');
  score_pos = score(:,2);
  auc = AUC(test_Y, score_pos');

• Standardization : the script applies zscore + mapstd, consistent with the paper; feel free to swap in your preferred preprocessing.

---

Quick test

addpath(genpath(pwd));
load('Data/DualMarker_demo.mat');

run('DualMarkerMainCode.m');      % produces geneRANK
run('KFoldCrossValidation.m');    % prints AUC results

---