We developed national machine-learning models to provide interpretable results, to predict concentrations of PM2.5 sulfate, nitrate, organic carbon (OC) and elemental carbon (EC) across the conterminous United States from 2005 to 2015 at the daily level.