Figure 4. Racial bias in datasets.

Demographic data (right) and receiver operating characteristic (ROC) curves (middle) from two brain proteomics datasets with differing representation of African American/black and non-Hispanic white participants in two different Alzheimer’s disease studies. In the dataset containing mostly white adults (top), amyloid precursor protein (APP) is a much better marker than heat-shock protein beta 1 (HSPB1), as shown by its larger area under the ROC curve in the top panel. In a more racially diverse dataset (bottom), the utility of HSPB1 as a potential biomarker for Alzheimer’s disease becomes more obvious. Reprinted with permission from Desaire et al.¹⁷ Copyright 2022, American Chemical Society.