This repository contains the code, SQL queries, and Jupyter notebooks used for the computational analysis of DTA mutational burden (DNMT3A, TET2, ASXL1) as a prognostic biomarker in myelodysplastic syndromes (MDS).
The project evaluates the impact of cumulative mutational burden on overall survival (OS) and leukemia-free survival (LFS) using both traditional survival methods and machine-learning models.
Mutations in DNMT3A, TET2, and ASXL1 (collectively known as DTA) are highly prevalent in:
- Age-related clonal hematopoiesis (ARCH / CHIP)
- Myelodysplastic syndromes (MDS)
- Myeloid neoplasms more broadly
While individual DTA mutations have been associated with CHIP biology and clonal evolution, the prognostic impact of cumulative DTA mutational burden in treatment-naïve MDS has not been fully explored.
This project evaluates whether the number and type of DTA variants contribute additional prognostic information beyond established clinical tools.
- SQL-based extraction of clinical, cytogenetic, and genomic features
- Harmonization of mutation calls across DTA loci
- Creation of cumulative DTA burden metrics:
n_dta(total DTA mutations)- gene-specific burden (e.g.,
n_asxl1,n_dnmt3a) - truncating vs. non-truncating subclassification
- Kaplan–Meier survival analysis
- Cox proportional hazards models
- Multiple comparison adjustments (Holm–Bonferroni)
- Bootstrap confidence intervals for C-index and IBS
- CoxNet (elastic net penalized Cox regression)
- Random Survival Forest (RSF)
- Gradient Boosting Survival Tree (GBST)
- SHAP explainability for feature attribution
- C-index with 95% bootstrap confidence intervals
- Integrated Brier Score (IBS) for calibration
- SHAP-based variable importance for interpretability
git clone git@github.com:pvhernandezlab/MDS_DTA_mutational_burden_study.git
cd MDS_DTA_mutational_burden_studyIf you have questions or run into any issues, please submit a GitHub issue with details of the cell output