Collaboration and Impressive Skill Mark First HuBMAP Kaggle Competition

Collaboration and Impressive Skill Mark First HuBMAP Kaggle Competition

May 24, 2021

Hacking the Kidney harnesses community effort to automate detection of microscopic kidney structure.

A combination of intense competition and expansive collaboration that reflected on HuBMAP’s own constitution marked Hacking the Kidney, what is likely to be the first but not the last Kaggle Competition sponsored by the Consortium. Held between Nov. 11, 2020, and May 10, 2021, the competition focused on mapping the human kidney at single-cell resolution by implementing a successful and robust machine-learning (ML) algorithm to automatically detect kidney glomeruli in three-dimensional microscopic images—a functional tissue unit, or FTU, detector.

The awards ceremony for the competition capped a three-day HuBMAP Annual All-Hands Meeting reviewing consortium members’ work and achievements over the past year.

HuBMAP “hope[s] to catalyze the development of an open global framework for comprehensively mapping the human body at cellular resolution,” said Katy Börner of Indiana University, one of the competition organizers and PI of the university’s HIVE mapping component. “So just like we used to have maps in large atlases but now we have Google Maps that we can run API queries against, we would like to have some of the anatomical atlases converted into a digital format that we can really run API queries against … Ultimately that means that we need to accelerate the development of the next generation of tools and techniques for constructing such a high-resolution, spatial-tissue-mapping atlas.”

Kaggle, the product of an eponymous Google subsidiary, is a platform designed for finding and publishing scientific datasets in a collaborative, Web-based fashion. By powering exploration and creation of models and ML tools and publishing an open collection of “Kaggle workbooks” that lay out these tools and solutions, Kaggle creates a working environment that leverages cooperation for scientific discovery. Kaggle Competitions serve to popularize the platform and help new generations of scientists and coders to learn and use it to solve real-world problems.

“The key behind the science and success of this competition was collaboration,” said competition support team member Leah Scherschel of Indiana University. “Kaggle is very unique in that it is part data science and part social media platform. Competitors … shared knowledge of the medical domain with one another because most of them only had machine-learning experience, not medical experience, and they showed each other how multiple AI frameworks such as TensorFlow, PyTorch and Keras could be applied to this challenge.”

On May 21, 2021, staff from HuBMAP and sponsors presented a number of awards to Hacking the Kidney competitors, distributing a total of $60,000 in prize money. The awards were:

Accuracy Prizes:

• 1st Place [$18,000] - Tom (tikutiku)
• 2nd Place [$10,000] - Team Gleb (bakeryproducts, kulyabin, and sherlockkay)
• 3rd Place [$3,000] - Team Whats goin on (shujun717 and rvslight)

Judges Prizes:

• Scientific Prize [$15,000] - Team DeepLive.exe (optimo, theoviel, and iafoss)
• Innovation Prize [$10,000] - Team deepflash2 (matjes, theudas, maddonix)
• Diversity Prize [$3,000] - Team 404! (Gavi_sr, Sophie, tpmeli, dskswu)

Kudos:

• Team Players! - Team Efforless Neuron (chienhsianghung, ulrich07, raphaelensae, faisalalsrheed, salimkhazem)
• Community Hero - Wojtek Rosa (wrrosa)
• Most Entertaining - Team deepflash2 (matjes, theudas, maddonix)

Diversity of background and approach proved key to the winning projects, as well as the competition in general.

“The challenge is that those images were such high-resolution that it was a problem even to look at them,” said support team member Marcos Novaes of Google. “So immediately some users jumped in sharing notebooks just on the visualization of the images with the right packages to decode those high-resolution TIFF images … It was amazing to see how users jumped in to help others in the forum and making comments on notebooks, and how that accelerated the competition.”

You can find write-ups, Kaggle workbooks, datasets and other information on the winning projects here.

You can watch a video recording of Hacking the Kidney’s awards ceremony here.

Hacking the Kidney was supported by sponsors Google HCLS, Roche, Deloitte, Deerfield Healthcare, Pistoia Alliance, CAS - American Chemical Society and Maven Wave, as well as the NIH Common Fund through the Office of Strategic Coordination/Office of the NIH Director under award OT2OD026671 and VU HuBMAP under award 1U54DK120058-01