We hail individual geniuses, but success in science comes through collaboration (Farrar, 2017). Biomedical breakthroughs come from collaboration that crosses boundaries. Boundaries created by disciplines, organizations, cultures, professions, and demographics. While cross-boundary collaboration in team science has demonstrated benefits, research also suggests they are unlikely to form, and when they do, are prone to coordination costs (Cooke & Hilton, 2015). This research project advances the Science of Team Science by understanding and assembling cross-boundary teams to conduct clinical and translational science. This research enables scholars and policymakers to design and assess dream teams. Toward the aim of understanding, we conduct archival studies of NIH's Clinical & Translational Science Pilot Grant programs at two institutions to reveal the team assembly factors that drive formation and success. Toward the aim of assembling, we leverage a newly developed team recommender system.The fundamental goal is to generate recommendations to shift the composition of teams submitting Pilot Grant applications in particular and cross-boundary scientific teams in general. Insights from this project come at a critical point in time, when cross-boundary science is essential for biomedical research. Toward that aim, this project has two key sources of intellectual merit. First, the research advances the science of team science (Cooke & Hilton, 2015), answering calls to better understand the multilevel determinants of science team success. How are they forming and which ones are performing? Following teams from formation to maturity provide a holistic understanding of the factors driving team formation and performance. Knowing how to better assemble scientific teams has the potential to improve the career productivity of a major source of human capital, and to hasten breakthroughs in biomedical research. Furthermore, not all cross-boundary teams are successful. Previous research provides competing advice for those assembling teams. On the one hand, cross- boundary teams are most likely to produce novel insights (Stvilia et al., 2011; Uzzi et al., 2013), but they are also most likely to suffer process losses from their coordination costs (Cummings & Kiesler, 2007; Cummings et al., 2013). This large scale analysis will allow us to discover patterns that differentiate the kinds of crossboundary teams who are ultimately successful from those who are not. The second source of intellectual merit is the development of a team recommender system for biomedical scientists to use to assemble cross-boundary teams for Pilot Projects. The Dream Team Builder was specifically mentioned in the NAS report as it builds on NIH-funded VIVO's ontology. We extend this tool to provide team recommendations, and study their effects on team assembly and future performance.