New Northeastern and UCD Seed-Funded Projects to Include Collaboration between PROTECT and UCD Researchers
Northeastern University and University College Dublin (UCD) recently announced seed funding for a project with the PROTECT and AI_PREMie teams to develop machine learning methods that identify and prioritize biological samples to train and test AI algorithms for disease detection. The project, titled “Sample Sleuth: Active Learning for Optimal Biomarker Sampling Strategies in Clinical AI,” is being funded alongside four other collaborative Northeastern and UCD research teams.
The AI_PREMie team, led by UCD Professor Patricia Maguire, has used biomedical, clinical, and machine-learning expertise to develop a prototype risk stratification tool to identify pregnant women with preeclampsia. Their techniques to develop the tool included the extraction of protein biomarkers from blood samples. These biomarker extractions then helped form the AI_PREMie predictive model that can diagnose and stratify the risk of preeclamptic toxemia (PET) in expectant mothers from as early as 20 weeks into their pregnancies.
PROTECT has a large inventory of participant samples that the AI_PREMie team’s biomarker extraction process could be used on. In an ideal world, the process would be used on every PROTECT blood sample to inform the development of other predictive models. However, extracting data at this scale is not feasible. The Sample Sleuth project will address this issue and make data extraction from PROTECT samples attainable. The collaborating researchers will adapt Active Learning techniques and develop an algorithm that can sift through the large number of PROTECT data and identify the samples that will be the most beneficial for biomarker extraction. By identifying the most beneficial samples, the scale of biomarker extraction will be far more practical and will provide researchers with data that can inform predictive models.
The Sample Sleuth project will be led by DMAC co-Lead Justin Manjourides, AI_PREMie Principal Investigator Patricia Maguire, UCD School of Computer Science Associate Professor Brian Mac Namee, and PROTECT Principal Investigator Akram Alshawabkeh. The project and Sample Sleuth algorithm will be impactful in reducing time and costs in situations where samples need to be identified for model training. Improved model training efficiency can then enable the development of more tools like AI_PREMie that can detect diseases and issues associated with pregnancy that threaten maternal and child health.
L to R: UCD Researchers Ana Le Chevillier, Brian Mac Namee, & Patricia Maguire with PROTECT researchers Justin Manjourides, Akram Alshawabkeh, & José Cordero