Program: Data Science & Artificial Intelligence (DSAI) - Final Year Project
Domain: Transcriptomics, Next-Generation Sequencing (NGS), Machine Learning
Tech Stack: R, edgeR, DESeq2, Random Forest, Decision Trees
Pancreatic Ductal Adenocarcinoma (PDAC) is highly aggressive and often diagnosed late. This project aims to identify reliable genomic biomarkers for PDAC by integrating a Next-Generation Sequencing (NGS) pipeline with advanced Machine Learning algorithms.
By analyzing RNA-seq count data, we extracted Differentially Expressed Genes (DEGs) and trained predictive models to classify cancer presence with high precision.
- Conducted initial mapping and quality control of raw sequencing data using FastQ Screen.
- Filtered low counts to ensure robust downstream statistical analysis.
- Utilized edgeR and DESeq2 libraries.
- Applied
rlog(regularized log) transformation to normalize raw gene expression data. - Filtered normalized counts using the Coefficient of Variation to isolate the most significant features.
- Split the processed genomic data into training and testing sets.
- Deployed Random Forest and Decision Tree algorithms to evaluate predictive performance based on the identified genetic biomarkers.
- Generated performance visualizations using ggplot2.
The Random Forest algorithm significantly outperformed the Decision Tree, proving to be a highly reliable architecture for this diagnostic classification task.
| Performance Metric | Random Forest Model | Decision Tree Model |
|---|---|---|
| Accuracy | 92.0% | 76.0% |
| Sensitivity | 96.0% | 64.0% |
| Specificity | 88.0% | 88.0% |
ML.R: The core R script containing the data normalization process, feature selection, and machine learning model training.Data files.rar: The genomic datasets utilized for training the models.- Final_Report.pdf:🧬 Biomarker Identification in Pancreatic Ductal Adenocarcinoma
Program: Data Science & Artificial Intelligence (DSAI) - Final Year Project
Domain: Transcriptomics, Next-Generation Sequencing (NGS), Machine Learning
Tech Stack: R, edgeR, DESeq2, Random Forest, Decision Trees
Pancreatic Ductal Adenocarcinoma (PDAC) is highly aggressive and often diagnosed late. This project aims to identify reliable genomic biomarkers for PDAC by integrating a Next-Generation Sequencing (NGS) pipeline with advanced Machine Learning algorithms.
By analyzing RNA-seq count data, we extracted Differentially Expressed Genes (DEGs) and trained predictive models to classify cancer presence with high precision.
- Conducted initial mapping and quality control of raw sequencing data using FastQ Screen.
- Filtered low counts to ensure robust downstream statistical analysis.
- Utilized edgeR and DESeq2 libraries.
- Applied
rlog(regularized log) transformation to normalize raw gene expression data. - Filtered normalized counts using the Coefficient of Variation to isolate the most significant features.
- Split the processed genomic data into training and testing sets.
- Deployed Random Forest and Decision Tree algorithms to evaluate predictive performance based on the identified genetic biomarkers.
- Generated performance visualizations using ggplot2.
The Random Forest algorithm significantly outperformed the Decision Tree, proving to be a highly reliable architecture for this diagnostic classification task.
| Performance Metric | Random Forest Model | Decision Tree Model |
|---|---|---|
| Accuracy | 92.0% | 76.0% |
| Sensitivity | 96.0% | 64.0% |
| Specificity | 88.0% | 88.0% |
ML.R: The core R script containing the data normalization process, feature selection, and machine learning model training.Data files.rar: The genomic datasets utilized for training the models.- Research Report: Download Final Report PDF : The comprehensive 34-page research thesis detailing the biological background, methodology, and full analytical results. (Clicking this link will directly download the PDF to bypass GitHub's preview limits).
The integration of Machine Learning with traditional NGS pipelines successfully identified key biomarkers for Pancreatic Ductal Adenocarcinoma with a 92% accuracy rate. This highlights the critical role of Data Science and AI in accelerating oncological diagnostics and precision medicine.