spell0/Batch-effects-leaderboard-data

Batch Effects Leaderboard Data

Data repository for the Batch Effects Leaderboard.

This repository contains the datasets used by the leaderboard application:

• massbench_adenocarcinoma
• massbench_alzheimer
• massbench_benchmark
• normal_tissue_878
• colon_3041

File organization

Each dataset is stored under:

data/datasets/<dataset_name>/

Typical files include:

• <dataset>_all.csv — preferred complete dataset for research workflows
• <dataset>_train.csv — original training split
• <dataset>_test.csv — original test split
• <dataset>_inference.csv — inference matrix
• <dataset>_predictions.csv — reference labels / prediction metadata used by the project

Recommended workflow

The Batch Effects Leaderboard app favors *_all.csv as the default labeled source.

Users may instead select *_train.csv.

The selected source can be evaluated using:

• train / validation / test batch splits
• train / validation only

The separate inference workflow uses an inference target file and does not require a test split.

Some samples may have missing labels. These samples can still be retained for unsupervised representation learning or batch-effect correction and should be excluded only from supervised loss and supervised metrics.

Project

These files are used by the Batch Effects Leaderboard research application.