Behavioral profile predicts ethanol preference in adolescent mice, but not in adults: A machine learning approach

Clinical studies suggest that adolescents display a complex behavioral profile characterized by traits that increase their susceptibility to alcohol experimentation and impaired control over use. In the present study, we applied a machine learning model to predict the impact of diverse behavioral phenotypes on ethanol preference during adolescence and adulthood in mice.

C57BL/6 and Swiss mice were assigned to one of two age groups: adolescents (starting at PN40) or adults (starting at PN120). Over the next 3 days, novelty-seeking, anxiety-like behavior, sociability, coping behavior, and natural reward response were evaluated using the following behavioral tests: hole-board, elevated plus maze, three-chamber sociability, forced swimming, and sucrose preference, respectively. During the subsequent 5 days, alcohol preference behavior was assessed using the two-bottle choice paradigm (10% ethanol). We trained machine learning regression models to predict alcohol preference in each age group.

The analysis model significantly predicted ethanol preference based on behavioral phenotypic profiles in mice during adolescence, but not in adulthood. Notably, the behavioral traits that contributed most to the prediction were sucrose preference and sociability time. Sucrose preference had a positive predictive value. Distinctively, sociability time had a negative predictive value, indicating an inverse relationship with ethanol preference.

These findings suggest that behavioral phenotypes during adolescence, particularly natural reward sensitivity and sociability, are key predictors of ethanol preference. The negative association between sociability and alcohol intake highlights the potential protective role of social interaction. The absence of predictive value in adulthood underscores adolescence as a critical developmental window during which behavioral traits may influence vulnerability to alcohol use.