< No: 32 >
2019


The HPA Kaggle Challenge

Based on the HPA Cell Atlas image collection, a computational competition was arranged to identify deep-learning solutions for classification of subcellular protein patterns. Challenges included training on highly imbalanced classes and predicting multiple labels per image. More than 2,000 teams participated, and the winning models far outperformed our previous model. These models can be used as classifiers to annotate new images, feature extractors, or pretrained networks for a wide range of biological applications.

Key publication



Figure legend: Opening page of the HPA Kaggle competition.


Key facts

  • 2,200 teams presented a diverse set of deep-learning solutions
  • The competition managed multilabel classification with proteins localized to several subcellular compartments
  • The top model can be used as a feature extractor to embed spatial localization in cellular models
  • The top-ranking models performed better than any previously published model