Demystifying Machine Learning With AutoML

A Nationwide Children’s Hospital team from the Office of Data Sciences recently won the Advanced ML tier in the precisionFDA Automated Machine Learning (AutoML) App-a-thon Challenge: Democratizing and Demystifying Artificial Intelligence (AI).

Each year, the FDA hosts challenges centered around data science and bioinformatics. This year, the challenge focuses on AutoML, a low-code ML technique that enables professionals without data sciences backgrounds to utilize ML.

According to the FDA, only about 15% of hospitals regularly use ML because of a lack of expertise. With the ultimate goal of making ML more accessible for health care organizations, the challenge brought experts from across the United States to the table to propose solutions that would support the application of AutoML to biomedical datasets.

“This challenge was an excellent fit for our newly formed Office of Data Sciences team to learn about an emerging AI technology, AutoML,” says Peter White, PhD, chief data sciences officer at Nationwide Children’s. “We envision this as a technology that could drive scientific discovery empowered by the Nationwide Children’s Data Lake, making machine learning more accessible to our researchers.”

The precisionFDA AutoML App-a-thon aimed to assess AutoML’s effectiveness in the biomedical field, particularly in data analysis and prediction. AutoML is designed to simplify and automate iterative tasks in the ML workflow, making it accessible to a broader range of users and expanding data analysis capabilities in health care.

The competition had two challenge tiers: an ML Tier for participants with limited data science experience and an Advanced ML Tier for participants with extensive data science experience. The Nationwide Children’s team competed in the Advanced ML tier, which included analyzing genomic data from a brain cancer study and an additional proteomic dataset from the National Cancer Institute Clinical Proteomics Tumor Analysis Consortium (NCI-CPTAC).

The team evaluated multiple AutoML technologies to identify the best approach for biomedical data analysis. They successfully developed an app that enables non-data science experts to input data and specify simple guidelines for the required model. The effectiveness of their application was demonstrated by creating a model that accurately predicted survival outcomes for brain cancer samples and identified sample swaps in the proteomics data.

“Competing alongside industry leaders in data science like IBM and winning is a testament to our team’s dedication and expertise. I am incredibly proud of their hard work and very grateful to our colleagues at the FDA for giving us this opportunity and for recognizing our entry in this way,” says Dr. White.