# Estimating Heterogeneous Treatment Effects in Breast Cancer with Multimodal TCGA Data

This repository contains an end-to-end research project on **causal machine learning** for oncology, focused on **heterogeneous treatment effects (hte)** using observational data. Using multimodal data from **tcga-brca**, the project estimates individualized treatment effects to support future personalized treatment strategies.

Full project report (pdf):
- `Estimating_Heterogeneous_Treatment_Effects_in_Breast_Cancer_Using_Multimodal_TCGA_Real_World_Data_Project_Report.pdf`

Scope
- Cancer type: breast cancer (tcga-brca)
- Modalities: clinical covariates + high-dimensional rna-seq gene expression
- Treatment setup: observational “treated vs untreated” framing (see report for operational definition)
- Goal: estimate individual / conditional treatment effects and evaluate decision policies

Methods
- Propensity score modelling + overlap diagnostics
- Covariate balance checks (smd / balance tables)
- HTE modelling
  - causal forests (grf-style)
  - meta-learners (s-, t-, x- learners)

Data Provenance & Ethics (TCGA / GDC)
This project uses data from the cancer genome atlas (tcga) program accessed via the **nci genomic data commons (gdc)**.
- Source: tcga-brca data retrieved from the gdc data portal / gdc api.
- Patient privacy: tcga data are de-identified; this repository does **not** distribute controlled-access data.

This repository is for research and educational purposes only. It is **not** a clinical decision tool and must not be used to guide patient care.