 17% of cancers have oncogenic drivers called abstract fusion oncoproteins, FOS, that form liquid-liquid phase separation, LLPS, biomolecular condensates, but the generality of this phenomenon is unknown. We tested 166 FOS in Hila cells and found that 58% formed condensates, displaying distinct physicochemical features and segregating into feature-based groups based on subcellular localization and biological function. Using machine learning, we developed a predictor of FO condensation behavior and discovered that 67% of 3000 additional first likely form condensates, with 35% predicted to function by altering gene expression. We also found a functional dichotomy between condensate positive and a negative FOS, with 47% of the predicted condensate negative FOS associated with cell signalling functions. Our datasets and reagents are rich resources for future research on FO condensation. This article was authored by Swarnendutrapathi, Hazheem Keshaneke, Scott D. Gorman and others. We are article.tv, links in the description below.