Discovering multi-omics latent factors linked to breast cancer survival by integrating RNA-seq, DNA methylation, CNV, and RPPA proteomics from TCGA-BRCA (n=636) using MOFA2.
- 15 latent factors discovered; Factors 1 + 2 explain 47% of coordinated variation across omics
- Factor 5 (anti-tumor immune response-PLA2G2D, TIGIT, ICOS, ZAP70): HR=0.63 [0.46–0.87], p=0.004 after age/stage adjustment
- Factor 7: +95.6 days restricted mean survival time (RMST) in high vs low group which is a "delayed protection" signal invisible to Cox models
- Factor 14: strongest RSF variable importance (3× next best), top risk factor
- Time-dependent AUC: 0.705 at 3 years; cross-method consistency (Cox, RSF, RMST, KM, TimeROC)
brca-navigator :- a Shiny app built with bs4Dash + plotly + visNetwork, deployed on shinyapps.io. Seven tabs:
| Tab | Purpose |
|---|---|
| The Story | Scrollytelling narrative of the analysis |
| Factor Explorer | Browse all 15 factors, variance, top weights, survival curves |
| Patient Map | UMAP embedding colored by factors, survival, PAM50 subtype |
| Gene Explorer | Search genes, view weights across factors, expression distributions, survival impact |
| Survival Simulator | Adjust factor scores interactively and see predicted survival curves |
| Analysis Flow | Visual pipeline diagram |
| About | Methods, reproducibility, references |
.
├── Data/ # Raw data (UCSC Xena download), MAE object
├── Data/GEO/ # GSE20685 for external validation
├── models/ # MOFA trained model (.hdf5 + .rds)
├── scripts/ # All pipeline R scripts
├── Results/ # CSVs: variance, survival, RSF, TimeROC, RMST, enrichment
├── logs/ # Preprocessing summaries, Cox results
├── figures/ # All publication-ready figures
├── dashboard_app/ # Shiny dashboard (app.R, data_prep.R, www/custom.css)
│ └── data/ # Dashboard-ready datasets
└── docs/ # Pipeline plan, analysis report, deployment notes
| Parameter | Value |
|---|---|
| MOFA latent factors | 15 |
| Random seeds | 42 (MOFA, RSF, control selection) |
| RNA top features | 8,000 |
| Methylation top features | 8,000 |
| Methylation imputation | KNN k=10, drop >20% missing |
| RSF trees | 1,000 |
| LASSO CV folds | 10 |
| TimeROC timepoints | 365 / 1095 / 1825 days |
| RMST tau | 1,825 days (5 years) |
| GEO validation | GSE20685 (Affymetrix GPL570) |
| Metric | Value |
|---|---|
| Patients | 636 (BRCA) |
| Omics integrated | 3 (RNA + Methylation + RPPA) |
| Factors discovered | 15 |
| Best protective factor (adjusted) | Factor 5, HR=0.63 [0.46–0.87], p=0.004 |
| Best risk factor (adjusted) | Factor 14, HR=1.44 [1.07–1.95], p=0.017 |
| RSF C-index | 0.594 |
| TimeROC AUC at 3 years | 0.705 |
| RMST best | Factor 7: +95.6 days (p=0.049) |
This work is licensed under CC BY-NC-ND 4.0. See LICENSE for details.
Copyright (c) 2026 Swati. All rights reserved.
TCGA data used in this project is available under TCGA publication guidelines.