Every year for a decade, the upcoming Vera Rubin Observatory (VRO) will witness the explosive deaths, or supernovae, of ~20,000 stars. Spectroscopy is needed to understand these transient events, but because these observations are resource intensive only a fraction of discovered explosions can be studied in detail. As a result, classification of these sources in the first few hours to prioritize follow-up will be critical for optimizing VRO’s scientific yield. This is a precarious task as it demands rapid inferences from sparse, noisy data. In this talk, I will summarize the ways we can use contextual information as a proxy for data from the event itself. I will first outline my methodology for consolidating supernovae and the photometric properties of their host galaxies into the GHOST database and analysis package. Next, I will describe our use of dimensionality reduction to identify host properties with the highest significance in distinguishing between transient classes. I will then report on our efforts to predict supernova class with only host galaxy information, and conclude by discussing the role GHOST will play in training transient software pipelines for VRO first light.