Unveiling Cognitive Processes in Mice through Reinforcement Learning Models in the FST and TST

Abstract

Simple behavioral tests like the forced swim test (FST) and tail suspension test (TST) are widely used to assess depression-like behaviors in rodents, primarily measuring immobility time. However, this approach oversimplifies behavioral readouts and overlooks the cognitive processes driving behavior, leaving the relationship between increased immobility and cognitive biases unclear. Here, we developed the SwimStruggleTracker (SST) to extract fine-grained behavioral trajectories and integrate computational modeling to methodically analyze behavior. Our findings reveal that behavior in the FST and TST follows reinforcement learning principles involving learning, consequence perception, and decision-making. Notably, the cognitive processes underlying behavior differ between the two tests, challenging the assumption that they are interchangeable for cross-validation. Regression analyses identify distinct behavior phases: early behavior is primarily influenced by learning-related factors, while later stages are more affected by consequence sensitivity. These findings suggest traditional analyses focusing final minutes may underestimate the role of learning and overemphasize consequence sensitivity. MotivationThe forced swim test (FST) and tail suspension test (TST) are among the most widely used paradigms for assessing depression-like behaviors in rodents. Yet, traditional analyses typically quantify only immobility during the final minutes, discarding rich temporal structure in the data and hindering efforts to uncover the cognitive mechanisms underlying these behaviors. To address this gap, we developed an automated tool that captures behavioral trajectories with fine temporal resolution and integrates computational modeling to dissect the cognitive processes driving behavior. Using this approach, we demonstrate that the FST and TST engage overlapping but partially distinct cognitive processes, and that the dominant cognitive components shift across different stages of the tests. HighlightsO_LISwimStruggleTracker (SST) accurately rejects passive movements, such as pendulum-like motion. C_LIO_LIReinforcement learning models capture the behavioral dynamics of mice in the FST and TST. C_LIO_LIDistinct winning models indicate that the FST and TST engage partially dissociable cognitive processes. C_LIO_LILearning factors dominate early stages, whereas consequence-sensitivity factors dominate later stages. C_LI