📻 Holiday Radio
Mele Kalikimaka
Welcome to the fifty fifth ever issue of Monday Morning Data Science from the Fred Hutch Data Science Laboratory. We are excited to show you what we have been working on (Fresh from the Lab), plus links that we think you would be interested in (Our Weekly Bookmarks Bar). Part of the purpose of this newsletter is to start conversations, so if you have a question or there is something you would like to share with us please let us know by responding directly to this email.
Fresh from the Lab
[How Biased Data and Algorithms Can Harm Health] This article from Johns Hopkins Bloomberg School of Public Health Magazine featuring the Data Science Lab’s Carrie Wright and Ava Hoffman highlights the issue of biased data and algorithms in healthcare, using the example of a machine learning model that could predict pain levels from osteoarthritis X-rays for white patients, but not for people of color. This bias was linked to the training data, which was based on physician reports that often under-recognized pain in marginalized groups. Adjusting the algorithm to consider patient-reported pain removed the racial bias. The article emphasizes that, contrary to hopes that AI could eliminate biases, it often perpetuates them due to biased data. Researchers are urged to understand the origins of these biases and recognize the limitations and biases in the data they use daily.
Our Weekly Bookmarks Bar
[Millions of new materials discovered with deep learning] Google's DeepMind has developed an AI tool, Graph Networks for Materials Exploration (GNoME), that has discovered 2.2 million new crystals, significantly accelerating the materials discovery process. Of these, 380,000 are highly stable and are potential candidates for experimental synthesis, which could enable the development of transformative technologies such as superconductors and next-generation batteries. External researchers have already independently created 736 of these predicted structures, demonstrating the effectiveness of the AI tool. GNoME's predictions have been made available to the research community, with the hope of driving forward research into inorganic crystals and demonstrating the potential of AI in materials discovery at scale.
As always you can contact us by replying directly to this email, you can contact the Data Science Lab at data@fredhutch.org, or you are welcome to join us on the Fred Hutch Data Slack Workspace. For more information about the Fred Hutch Data Science Lab, visit our website: https://hutchdatascience.org/. See you next week!
- The Fred Hutch Data Science Laboratory