Transcriptomic Data Integration for Machine Learning

Cross-platform transcriptomic integration workflow for biomarker discovery and machine learning applications

---

📖 Overview

This project provides a reproducible workflow for integrating transcriptomic datasets generated from different studies and technologies (Microarrays and RNA-seq) with the goal of generating machine-learning-ready datasets.

The workflow focuses on reducing technical variability while preserving biological signals relevant to CDK inhibitor resistance studies.

The repository is structured as a tutorial-style analysis designed to explain each preprocessing step rather than simply executing a black-box pipeline.

---

🎯 Objectives

• Integrate transcriptomic datasets from independent studies
• Harmonize data from different experimental platforms
• Reduce technical variability and batch effects
• Preserve biological information relevant to drug resistance
• Generate clean datasets suitable for downstream machine learning

---

🛠 Technologies and Packages

Programming

• R

Main packages

• tidyverse
• limma
• sva
• edgeR
• AnnotationDbi
• preprocessCore
• biomaRt
• ggplot2

---

🔬 Workflow

Microarray workflow

1. Probe annotation
2. Probe summarization
3. Quantile normalization
4. Log2 transformation
5. Low-expression filtering
6. Cross-study gene matching
7. Batch effect correction (ComBat / SVA)
8. Exploratory quality control
9. ML-ready matrix generation

RNA-seq workflow

1. Gene annotation
2. Count summarization
3. Low-expression filtering
4. TMM normalization
5. Log transformation
6. Cross-study integration
7. Batch correction
8. Exploratory PCA analysis
9. ML-ready matrix generation

---

📂 Repository Structure

.
├── Microarrays_data
│   ├── merging_samples.Rmd
│   └── index.html
│
├── RNA-seq_data
│   ├── merging_samples_rnaseq.Rmd
│   └── index.html
│
├── index.html
└── README.md

---

📊 Key concepts

• Multi-cohort transcriptomic integration
• Cross-platform harmonization
• Feature engineering
• Batch effect correction
• Exploratory data analysis
• Reproducible workflows
• Machine learning preprocessing

---

🚀 Outputs

The final outputs generated by this workflow include:

• Annotated expression matrices
• Normalized datasets
• Batch-corrected datasets
• Exploratory QC visualizations
• ML-ready matrices for downstream predictive modeling

---

📚 Notes

This repository was designed as a guided and educational workflow intended to explain the rationale behind each preprocessing step. The focus is on interpretability and reproducibility rather than only providing executable scripts.