Identity Domains: A Computational Framework for Personality Analysis

Post by Deborah Joye

What's the science?

Personality is the collection of individual behaviors or traits that differ from one human to the next. The ability to organize individual differences in behavior into distinct categories is important for understanding the biological underpinnings of both healthy and pathological behaviors. In humans, many individual differences have been categorized by psychologists into varying personality traits, resulting in the widespread use of personality tests to determine the trait make-up of individuals. However, personality tests tend to rely on self-report questionnaires and do not track actual behaviors as they occur. Like humans, mice also exhibit individual differences in their behaviors. Some mice prefer to stay close to the nest, whereas others leave the nest to explore the environment. Some mice will readily approach a stranger, while others are more reserved. Organizing individual differences amongst other species has been challenging since there are few conceptual frameworks with which to comprehensively categorize behaviors into consistent traits. This week in Nature Neuroscience, Forkosh, Karamihalev, and colleagues present a computational framework that organizes individual behaviors into trait-like dimensions that are stable across development, consistent across social settings, and correlated with gene expression differences within the brain.

How did they do it?

The authors built a semi-naturalistic arena and filled it with different features for mice to interact with such as ramps, feeders, dividing walls, and hiding places. The authors then tracked the behavior of individual mice as they interacted with their cage-mates and environment over several days. They classified both individual behaviors, like movement and foraging and interactions between the mice, like dominance behaviors and other social contacts. The authors then trained their linear discriminant analysis algorithm to look for 60 unique behaviors within the behavioral dataset. Their algorithm was specifically designed to isolate dimensions of the dataset that are the best at discriminating one mouse from another, which they called identity domains. The algorithm does this by maximizing trait variability between each mouse, while also maximizing trait consistency within one mouse. To ensure that their algorithm functioned as planned, the authors validated the analysis on two separate groups of mice and found consistent results. To determine if identity domains were consistent within mice across development, the authors profiled identity domains of juvenile mice, then profiled the same mice as adults. To test whether identity domains remained consistent in different social settings, the authors profiled groups of mice, then mixed the mice up into different groups and profiled them again. Finally, to determine whether identity domains correlated with changing gene expression in the brain, the authors performed RNA sequencing three brain regions (basolateral amygdala, insular cortex, and medial prefrontal cortex) of profiled mice.

What did they find?

The authors’ algorithm captured four identity domains – consistent dimensions of the data that described the stable behavior of individual mice over time. Interestingly, the authors found that when mice were profiled as both juveniles and adults, three of the four assigned identity domains remained stable, suggesting that identity domains capture traits that are consistent across development. The authors then mixed up groups after they had been profiled and found that while mice changed some behaviors in new social settings, their assigned identity domains remained stable. Using RNA sequencing, the authors demonstrated that gene expression variability in 3 different regions could be predicted using identity domain scores, suggesting that behavioral differences captured by identity domains and gene expression in the brain are associated. Finally, the authors investigated the identity domains of mice with known behavioral phenotypes, such as mice that are known to exhibit high anxiety behavior. The authors found that their assigned identity domain scores were highly associated with expected personality traits, suggesting real-world relevance of the identity domain scores.

What's the impact?

Individual differences in behavior are quite difficult to study. This work presents a novel framework that offers a more objective study of personality by tracking real behavioral output and categorizing it into trait-like identity domains in mice. Interestingly, identity domains capture differences that are stable over time and in different social contexts. Moreover, the correlation between identity domain scores and gene expression differences in several brain areas suggests that this tool can capture stable behavioral differences that are reflective of fundamental differences in brain function.

Forkosh et al., Identity domains capture individual differences from across the behavioral repertoire, Nature Neuroscience (2019). Access the original scientific publication here.