GlncDDR: Predicting DNA Damage Response-Associated lncRNAs Using Graph-Based Machine Learning

GlncDDR is a machine learning pipeline designed to predict long non-coding RNAs (lncRNAs) involved in the DNA Damage Response (DDR). This pipeline leverages graph-based node embeddings (Node2Vec) and gene expression data from The Cancer Genome Atlas (TCGA) to uncover novel DDR-associated lncRNAs.

---

Key Features

• Graph-based learning using Node2Vec over WGCNA co-expression networks
• Node2Vec was implemented(node2vec: Scalable Feature Learning for Networks (KDD 2016) )
• Classical ML models: Logistic Regression, Random Forest, Support Vector Machine
• Trained on known DDR/non-DDR genes from literature
• Predicts and ranks DDR-associated lncRNAs from GENCODE v36
• Model performance was assessed using five-fold cross-validation on labeled DDR and non-DDR gene sets.

---

Folder Structure

GlncDDR_Complete_Pipeline/
├── requirements.txt
├── main.py
├── embeddings/
│   ├── train_emb_len100.csv
│   ├── test_emb_len100.csv
│   ├── lncrna_emb_len100.csv
│   └── protein_emb_len100.csv
├── ml_pipelines/
│   ├── embedding/
│   │   └── run_embedding.py
│   └── ml/
│       ├── training.py
│       ├── testing.py
│       ├── predict_lnc.py
│       ├── predict_prot.py
│       └── utils.py
└── output_dir/

Data Availability

Note: The raw expression data files used to generate graph-based embeddings (e.g., TCGA expression profiles) are not included in this repository due to size constraints (multiple terabytes).

To access the raw data or reproduce the full embedding pipeline from scratch, please contact:

Snehal Shah snehals@clemson.edu Liangjiang Wang liangjw@clemson.edu

---

How it works

1. Install Dependencies

Recommended: conda

conda create -n glncddr python=3.9 -y \
conda activate glncddr

Install Python deps

pip install -r requirements.txt

2. Running the ML scripts

Step 1: Run Embedding (optional, one-time only)

Use this step if you’re starting from raw gene expression:

python ml_pipeline/embedding/run_embedding.py \
--input data/train.csv \
--output_dir embeddings \
--vector-size 100 \
--walks 5 \
--length 10

*Repeat for: test.csv, lncrna.csv, protein.csv

Step 2: Train and Predict

python main.py \
--train embeddings/train_emb_len100.csv \
--test embeddings/test_emb_len100.csv \
--predict_lnc embeddings/lncrna_emb_len100.csv \
--predict_prot embeddings/protein_emb_len100.csv \
--output output_dir/

Output Files

File	Description
training_metrics.txt	Training set performance metrics (Accuracy, Sensitivity, etc.)
test_metrics.txt	Test set evaluation results
combined_roc_pr.png	ROC and PR curves from cross-validation
*.xlsx	Ranked predictions for lncRNAs and protein-coding genes. Each file consists of three sheets - lr_predict_proba, rf_predict_proba and svm_predict_proba

Example Performance Table

Model	Accuracy	Sensitivity	Specificity	MCC	F1 Score	ROC-AUC
LR	0.85	0.79	0.94	0.72	0.86	0.95
RF	0.86	0.81	0.94	0.73	0.87	0.95
SVM	0.82	0.72	0.970	0.68	0.83	0.95

Example Prediction File

Ensembl	Genes	predict_proba
ENSG00000207744	MIR10B	1
ENSG00000264171	MIR4305	1
ENSG00000227125	AP002856.1	1
ENSG00000234277	LINC01641	1
ENSG00000231510	LINC02782	1