| Author | Shivani Malik |
|---|---|
| Date | 2025-06-20 |
| Title | SKCM README |
This project identifies and validates prognostic biomarker sets in skin cutaneous melanoma (SKCM) using statistical and machine learning approaches.
Malik S, Tomer R, Arora A and Raghava GPS (2026) Identification of multiple prognostic biomarker sets for risk stratification in SKCM. Front. Bioinform. 5:1624329. doi: 10.3389/fbinf.2025.1624329
https://doi.org/10.5281/zenodo.19881489
- Data Retrieval: Gene expression data (473 samples, 60,660 genes)
was downloaded from TCGA using
TCGAbiolinksin R. - Normalization: Data was normalized to TPM.
- Statistical Analysis: Performed correlation analysis, survival analysis (Cox, Kaplan-Meier), and LASSO.
- Survival Class Creation: OS time was used to create 4 survival classes (0--1, 1--3, 3--5, >5 years → Class 0,1,2,3).
- Feature Selection: Applied SVC-L1, RFE, SelectKBest, and SFS to identify top genes.
- Class Balancing: Used SMOTE to balance the dataset.
- Biomarker Sets: Created 7 biomarker sets (each with 20 genes).
- Model Performance: AUC ranged from 0.84 to 0.91, and Kappa from 0.48 to 0.64 across sets.
- Model Training: Applied many models but best Trained classifiers (CatBoost) and evaluated using AUC-ROC.
- Validation: Validated on GSE65904 (214 samples) using
overlapping genes.
- First Set: 15 matching genes → AUC 0.85 (train), 0.83 (test)
- Third Set: 12 matching genes → AUC 0.85 (train), 0.86 (test)
This repository contains all relevant files and scripts used to identify and validate prognostic biomarkers in SKCM using statistical and machine learning approaches.
Contains lists of genes identified through feature selection and validated on a new dataset.
-
Primary_biomarker/List_1st_set.txt– First set of 20 genes selected from TCGA using SVC-L1.List_2nd_set.txt– Second set of 20 genes selected using alternative feature selection methods.List_3rd_set.txt– Third set of 20 genes selected from TCGA.List_4th_set.txt– Fourth set of 20 selected biomarker genes.List_5th_set.txt– Fifth set of 20 selected biomarker genes.List_6th_set.txt– Sixth set of 20 selected biomarker genes.List_7th_set.txt– Seventh set of 20 selected biomarker genes.
-
Validated_biomarker/List_match_1st.txt– List of 15 matched genes from the first biomarker set validated on GEO dataset (GSE65904).List_match_3rd.txt– List of 12 matched genes from the third biomarker set validated on GEO dataset (GSE65904).
Includes scripts and data access instructions for both primary (TCGA) and validation (GEO) datasets.
Script/TCGA_SKCM.R– R script for downloading SKCM expression data usingTCGAbiolinks.
Link/Data_link.txt– Contains links to TCGA data portal and UCSC Xena browser for direct download.
Script/GEO_SKCM.R– R script for downloading GSE65904 dataset.
Link/Data_link.txt– Contains GEO accession links.
Includes preprocessing, feature selection, and machine learning implementation scripts.
-
Preprocessing_TCGA.py– Preprocessing steps: Remove 50% zeroes and low variance gene. -
Feature_selection_ML_TCGA.py– Feature selection using SVC-L1 on the basis of ranking, ML training and evaluate performance -
15_1stset_ML_GEO.py– ML analysis on 15 matched genes from the primary biomarker set. -
12_3rdset_ML_GEO.py– ML analysis on 12 matched genes from the third biomarker set.
📁 TCGA Folder Included 🧬 gene_data.csv Contains gene expression values for all samples. Rows represent samples and columns represent genes. Usage: Used as input features for downstream analysis.
🏥 TCGA-SKCM_clinical.csv Clinical metadata of patients, including: Age, Gender, Tumor stage, Survival time, vital status Usage: Used to incorporate clinical variables and perform correlation with gene expression and survival outcomes. Also used as features for ML analysis.
⏳ OS_time_in_days.csv Contains the overall survival (OS) time, calculated in days, for each patient. Usage: Used for survival analysis, including prognostic modeling and outcome evaluation.
🔗 gene_clin_all_features.csv Integrated dataset combining: Gene expression data and Clinical features Usage: Used as the final feature matrix for training and evaluating machine learning models.
📊 gene_clin_OS_category.csv Dataset with categorized survival labels (e.g., 0–1 year, 1–3 years, 3-5, >5 years), used for multi-class classification tasks. Usage: Used as the target variable for multi-class survival prediction models. 📁 ML_data/primary_set_data/
This folder contains the processed dataset used for machine learning after feature selection.
📊 Description: Includes training and testing datasets Contains expression values of 20 selected genes Features were selected using the SVC-L1
GEO/GSE65904/ This folder contains gene expression and clinical data from the GSE65904 dataset.
📂 Data/ 🧬 gene_annotation.csv Contains mapping information for gene identifiers (e.g., probe IDs to gene symbols). Usage: Used to convert probe-level data into gene-level expression data for analysis.
📊 geo_expression_data.csv Contains gene expression values for all samples in the GEO dataset. Usage: Used as the validation dataset to test model performance.
🧾 geo_phenotype_data.csv Contains sample metadata, including clinical and phenotypic information. Usage: Used to link expression data with clinical outcomes for validation.
📁 GEO/GSE65904/ML_13_3rd/ This folder contains the processed subset of the GEO dataset used for machine learning validation after preprocessing and feature alignment.
📂 preprocessing/ 🧬 13_3rdset_expression_with_DSS.csv Contains gene expression data along with DSS (Disease-Specific Survival) information for selected samples. Usage: Used to integrate survival outcomes with expression data for validation of prognostic models.
📊 geo_final_13.csv Final processed dataset containing selected features aligned with the training dataset (e.g., selected genes). Usage: Used as the external validation dataset for testing trained machine learning models on independent data.
📁 GEO/GSE65904/ML_15_1st/ This folder contains a processed subset of the GEO dataset based on a 15-gene feature set, used for validation of machine learning models.
📂 Preprocessing data/ 🧬 15_expression_with_DSS.csv Contains gene expression data for the selected 15 genes along with DSS (Disease-Specific Survival) information. Usage: Used to associate gene expression with survival outcomes for prognostic validation.
📊 final_expression_matrix.csv Processed gene expression matrix after normalization and filtering. Usage: Used as the base dataset for feature alignment and downstream validation.
🔗 final_expression_with_DSS.csv Integrated dataset combining gene expression values with DSS information. Usage: Used for survival-based validation and model evaluation.
📁 Main File 📊 geo_final_15.csv Final dataset with selected 15-gene features aligned with the training dataset. Usage: Used as an independent validation dataset to evaluate trained machine learning models.
All the GEO and TCGA relevant data files and models could be downloaded from the google drive link: https://drive.google.com/drive/folders/1HIVw_rfpHolB3tAvJTtOZVV0USOANil-.