Meta-Album PRT Mini: Subcellular Human Protein Images

The Human Protein Atlas project provides 840 RGB images of human cells, each resized to 128x128 pixels. This subset, curated for Meta-Album, contains images labeled with one of 21 subcellular protein localization classes. It was created by Felix Mohr in June 2022 from a larger multi-label Kaggle competition dataset.

Use Cases

• Training multi-class image classifiers based on protein organelle localization labels.
• Benchmarking few-shot or meta-learning algorithms on a standardized microscopy dataset.
• Developing computer vision models for biological pattern recognition in human cells.

Strengths

• Contains 840 labeled images across 21 distinct protein localization classes.
• Images are standardized to a consistent 128x128 pixel resolution.
• Derived from the well-cited Human Protein Atlas, a recognized source in proteomics.

Limitations

• The dataset is a curated subset; original multi-label information and images with multiple classes were removed.
• Image resolution has been reduced from the original 512x512 pixels, which may limit detail for some analyses.
• Row and column counts for the underlying data files are unknown from the provided metadata.

Provenance

Source

The Human Protein Atlas (proteinatlas.org) and a related Kaggle competition.

Collection Method

Original RGBY images were converted to RGB by dropping the yellow channel, filtered to single-class images, and resized.