Item

Abstract

Animal behavior contains rich structure across many timescales, but there is a dearth of methods for characteriz- ing the acquisition and improvement of task competence over the long run. Inspired by the goals and techniques of genome-wide association studies, we report a data-driven method–the choice-wide behavioral association study: CBAS–that systematically identifies differences between the choices of groups of subjects. CBAS compares the frequencies of many sequences of choices between two groups, then uses powerful, resampling-based, multiple comparisons methods to identify the sequences that differ significantly between the groups. We illustrate CBAS by comparing the behavior of wild-type (WT) to genetically modified rats and two structurally different reinforcement-learning (RL)-based computational agents as they learn a series of contingencies in a spatial alternation task. Each contingency requires subjects to visit arms of a track in a repeating pattern. When we apply CBAS to all se- quences up to six arm visits long made by two RL agents that learn with alternative strategies, it identifies distinct sequences that can be related to the different algorithms used by the agents. When we use CBAS to compare WT rats to those haploinsufficient for a high-confidence autism spectrum disorder risk gene (Scn2a), it identifies specific and consistent ways that Scn2a haploinsufficient rats differ throughout all phases of learning. Through identifying the choices that differ between groups of subjects, CBAS provides a uniquely informative framework to interpret neural function and its changes in the context of disease processes.