Extracting the Dynamics of Behavior in Decision-Making Experiments

Abstract

Understanding how animals update their decision-making behavior over time is an important problem in neuroscience. Decision-making strategies evolve over the course of learning, and continue to vary even in well-trained animals. However, the standard suite of behavioral analysis tools is ill-equipped to capture the dynamics of these strategies. Here, we present a flexible method for characterizing time-varying behavior during decision-making experiments. We show that it successfully captures trial-to-trial changes in an animal's sensitivity to not only task-relevant stimuli, but also task-irrelevant covariates such as choice, reward, and stimulus history. We use this method to derive insights from training data collected in mice, rats, and human subjects performing auditory discrimination and visual detection tasks. With this approach, we uncover the detailed evolution of an animal's strategy during learning, including adaptation to time-varying task statistics, suppression of sub-optimal strategies, and shared behavioral dynamics between subjects within an experimental population. 45 et al., 2009). Other work from Kattner et al. (2017) extended the standard psychometric curve to allow its parameters to vary continuously across trials. Bak et al. (2016) described a model for defining smoothly evolving weights that could track changing sensitivities to specific behavioral covariates. While the model could track behavioral dynamics in theory, the optimization procedure strongly constrained both the complexity of the model and the size of the data to which it could 50 65 unprecedented insight into the development of behavioral strategies. ResultsOur primary contribution is a method for characterizing the evolution of animal decision-making behavior on a trial-to-trial basis. Our approach consists of a dynamic Bernoulli generalized linear model (GLM), defined by a set of smoothly evolving psychophysical weights. These weights 70 characterize the animal's decision-making strategy at each trial in terms of a linear combination of available task variables. The larger the magnitude of a particular weight, the more the animal's decision relies on the corresponding task variable. Learning to perform a new task therefore involves driving the weights on "relevant" variables (e.g., sensory stimuli) to large values, while driving weights on irrelevant variables (e.g., bias, choice history) toward zero. However, classical 75 modeling approaches require that weights remain constant over long blocks of trials, which precludes tracking of trial-to-trial behavioral changes that arise during learning and in non-stationary task environments. Below, we describe our modeling approach in more detail. Dynamic Psychophysical Model for Decision-Making TasksAlthough our method is applicable to any binary decision-making task, for concreteness we intro-80 duce our method in the context of the task used by the International Brain Lab (IBL) (illustrated in Figure 1A) (IBL et al., 2020). In this visual detection task, a mouse is positioned in f...

Mice alternate between discrete strategies during perceptual decision-making

“…One could imagine, for example, that a continuous state governing the animal’s degree of engagement drifts gradually over time, and that the GLM-HMM simply divides these continuous changes into discrete clusters. To address this possibility, we fit the data with PsyTrack, a psychophysical model with continuous latent states [55, 56]. The PsyTrack model describes sensory decision-making using an identical Bernoulli GLM, but with dynamic weights that drift according to a Gaussian random walk (see Methods sec.…”

“… (a) Cross-validation performance of the 3 state GLM-HMM compared to PsyTrack [55, 56] for all 37 mice studied (each individual line is a separate mouse; black is the mean across animals). Test loglikelihood is considerably higher for all mice for the GLM-HMM than for PsyTrack, which assumes that the mouse uses a set of smoothly evolving GLM weights to make its decision at each trial.…”

“…For example, one could imagine a mouse's attention gradually increasing or decreasing over time, rather than changing suddenly. In order to address this possibility, we performed model comparison with the PsyTrack model of [39,40]. PsyTrack assumes that an animal makes its choice according to…”

“…Specifically, the animal is assumed to use a set of slowly changing weights to make its decision on each trial. This model is typically used to characterize the learning behavior of mice [39,40], but we here apply it to model stationary decision-making. While we are able to successfully recover PsyTrack as the best fitting model if we generate choice data from PsyTrack ( Fig.…”

Continuous multiplexed population representations of task context in the mouse primary visual cortex

“…Our results show that V1 activity is systematically modulated by the task the animal is engaged in. Recent behavioral explorations have demonstrated that the rule that the animal bases its decisions on can change during a behavioral session, even when the task structure remains unchanged (Roy et al, 2020). It will be important to see whether such spontaneous changes in behavioral strategy are reflected in V1 activity in a similar manner to how changes in task context are represented.…”