Particulate matter (PM) emissions from spark-ignited engines, particularly those with direct injection, now exceed those of light-duty diesel engines equipped with diesel particle filters. Fuel chemistry is one of several interacting factors that determine the amount of PM produced during combustion. Understanding the relationship between fuel chemistry and PM emissions is therefore essential in identifying fuels with a lower tendency to form soot. However, existing predictive models have been shown to perform poorly when fuel blends include a large proportion of oxygenated molecules. In this study, we report a new analysis of the data from the EPAct V2 data set with the objective of developing an emission index that adequately handles the effects of oxygenate blending on vehicle and engine emissions. Our approach uses regularized linear regression to select the most important parameters in predicting normalized vehicle emissions as a function of fuel composition and bulk fuel properties. Interestingly, our data-derived metric reproduces a similar functional form to the particulate matter index equation, including a chemical tendency to form soot as well as the vapor pressure. The resulting metric better predicts PM formation in the EPAct V2 data set as a function of fuel properties and composition than existing models.