Meeting Abstract

P1-119    A hybrid, deep-learning pipeline for social network analysis in bumblebee colonies Ruttenberg, DM*; Wolf, SW; Knapp, D; Webb, AE; Kane, A; Gee, T; Breinyn, I; LeChance, J; Cohen, D; Kocher, SD; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University; Princeton University dmr4@princeton.edu http://scholar.google.com/citations?user=5qxgrbwAAAAJ&hl=en&oi=sra

How social colonies emerge from a set of individuals is an elusive question in evolutionary biology. Answering this requires studying not just the suite of social individuals, but also understanding how social networks are formed at the colony level and how individual variation structures those emergent properties. Bumblebee (Bombus impatiens) colonies represent an excellent, perturbable model for studying these dynamics, due to their small colony size, well-studied genetics, simple communication networks, and annual life history. We developed a hybrid pipeline integrating the robustness of tag-based tracking systems (ArUCO) with state-of-the-art deep-learning tools (SLEAP) to study the proximity network of social interactions of bumblebees in both a whole-colony setting and in small, queenless groups. We show that bumblebee social networks are structured by “influencer” bees, and that the presence of these influencers cannot be explained by spatial fidelity alone. Moreover, by comparing our results to a theoretical “null” model, we characterize several of the hallmarks of the bumblebee social network that can be used as emergent, colony-level phenotypes in future studies.