Output Dataset: Processed Adversarial Sensitivity Maps and Trained Models Derived from TCG

Harvard Dataverse hosts processed outputs from a multimodal adversarial sensitivity analysis of Triple Negative Breast Cancer (TNBC). The dataset includes trained image and gene expression classifiers, adversarial sensitivity maps, batch-corrected expression matrices, and analysis results derived from public TCGA, GEO, BreakHis, and METABRIC sources. Author Koushik Chowdhury last updated the dataset on April 13, 2026.

Use Cases

• Benchmarking model robustness based on adversarial attacks (FGSM, PGD) mentioned in the description
• Analyzing pathway enrichment in TNBC based on the statistical analysis results included
• Training or fine-tuning histopathology image classifiers based on the provided ResNet50 and EfficientNet models
• Investigating gene expression patterns in TNBC based on the ComBat-corrected matrices and Random Forest/XGBoost classifiers
• Correlating adversarial sensitivity with patient survival based on the included survival data

Strengths

• Includes five trained histopathology image classifiers and two gene expression classifiers
• Contains per-patient adversarial sensitivity maps for two specific TCGA-BRCA cohorts (A2 and E2)
• Derived from multiple established public sources (TCGA, GEO, BreakHis, METABRIC)

Limitations

• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• Raw image and expression data are not included and must be obtained from original sources

Provenance

Source

Processed outputs derived from TCGA, GEO, BreakHis, and METABRIC datasets.

Collection Method

Generated via multimodal adversarial sensitivity analysis using FGSM and PGD attacks.

Time Range

null

Freshness

Last updated 2026-04-13 16:28:56; freshness should be verified

Geography

null