hide
Free keywords:
-
Abstract:
Behavior forms the foundation upon which we interpret neurophysiology and model neuropsychiatric disease in animals. However, we have a limited understanding of the causes underlying the complex and dynamic decision-making processes used by animals. Although it is common to use hypothesis driven or model-based approaches to identify differences in behavior, those approaches are often based on simple and unverifiable assumptions. Furthermore, standard methods for analyzing behavior can often be too coarse to provide information about statistically subtle differences that occur during learning. Therefore, we present the choice-wide behavioral association study (CBAS) as a new approach for analyzing group differences in learning. This is motivated by data-driven approaches prevalent in genomics, such as the genome-wide association study (GWAS), which solved related problems in genetics. CBAS breaks down behavior into shorter sequences of choices, and then uses powerful, resampling-based, multiple comparison corrections to identify choices that differ in prevalence between groups of animals. First, we apply CBAS to a set of structurally different Reinforcement Learning (RL) agents; CBAS discovers many, interpretable, sequences of choices that distinguish the agents. Then, we apply CBAS to a cohort of 240 rats to discern the effect of Scn2a haploinsufficiency (Scn2a+/-) on the learning of a succession of spatial alternation contingencies. Scn2a expresses a neuronal sodium channel, and its disruption has been strongly associated with autism spectrum disorder (ASD). CBAS finds many choice sequences that differ between Scn2a+/- rats and their wild-type littermates, revealing a persistence of choices related to the prior contingency. This indicates that Scn2a+/- rats are slower to transition from a prior state of knowledge to current goals. Through developing a data-driven behavioral analysis, CBAS, we can richly phenotype the effect of a high-risk ASD gene.
