Neurocognitive Joint Models Symposium

Within model-based neuroscience, a major goal is understanding how brain-behavioral relationships unfold in a dynamic manner. Toward this end, sequential sampling models have been a major theoretical engine via which links between cognitive processes underlying perceptual decisions and neural signatures have been modeled trial-to-trial. These single-trial neurocognitive joint models have demonstrated advantages over more traditional methods of analysis in neuroscience, including the ability to partition variance in the neural signal that is/is not attributable to circumscribed cognitive processes. Although this attribute holds great potential for the study of populations with high intra-individual neural and cognitive variability (e.g., psychiatric populations), to date, this work has focused almost entirely on “normative” populations. This talk will test the robustness and translational utility of such models. First, we test the performance and feasibility of EEG-based single-trial joint models of social perceptual decisions in two clinical samples—schizophrenia and bipolar disorder—relative to more traditional methods. Second, we will address the clinical translational utility of such methods, assessing their ability to capture and arbitrate between various neural explanations of cognitive disruptions observed in patient populations.

Model-based cognitive neuroscience integrates cognitive models with brain data to model mechanistic connections between neural computations and the cognitive processes they purport to underlie. Recent methodological advancements have enabled such modeling to occur at the trial level, modeling momentary variations in cognitive processes and linked brain states. This has led to major advances in our theoretical understanding of brain-behavior links and demonstrated considerable benefits over traditional methods of analysis in neuroscience. This symposium brings together cutting-edge research on neurocognitive joint modeling to refine our understanding of brain-behavior relationships across diverse domains and the methodologies developed for these advancements. Talks will begin with foundational methodological research questions in joint modeling, including methods for refining linking functions (Turner; fMRI) and the pros/cons of different model architectures (Nunez; EEG). Subsequent talks will cover how various applications of joint modeling can be used to advance our understanding of decision-making across multiple domains and populations. This will include how neural and oculomotor signals support decisions across perceptual/value-based contexts (Fernandez; EEG/eye-tracking); subcortical contributions to decision urgency and reward processing (Miletić; fMRI); and testing the feasibility and translational utility of joint modeling in psychiatric disorders, including schizophrenia and bipolar disorder (Lasagna; EEG). This symposium will provide a comprehensive overview of recent methodological advances in neurocognitive joint modeling, offering new perspectives on how these approaches can refine our understanding of decision-making in both typical and clinical populations.

Model-based Cognitive Neuroscience is a field that seeks to understand, and benefit from, connections between functional brain recordings and cognitive models. Many researchers in this field have used extracted trial-by-trial EEG measures within models to help answer various research questions about (neuro-)cognition. For instance, how do Event-Related Potentials encode evidence accumulation during decision-making? However, different types of modeling connections of cognitive parameters to single-trial EEG measures have been used, and it is not clear what the costs and benefits are of each type of modeling connection. These include so-called “directed” models where single-trial EEG measures are, essentially, independent variables in a cognitive model that describes participant behavior in experimental tasks. These also include so-called “integrative” models where cognitive model parameters describe both task behavior and EEG measures as dependent variables on every trial. Currently, there is little insight into the best model structure to answer empirical questions using real data. It is widely expected that single-trial EEG will have contaminant data from processes that are not explicitly modeled. But how then does the contaminant data draw unwanted conclusions in both “directed” and “integrative” models? We present work that aims to clarify the best neurocognitive modeling structures for answering empirical questions.

A prominent problem in the field of model-based cognitive neuroscience is deciding which brain areas are related to which cognitive construct. Although early work allowed for agnostic specifications for how these links might be structured, as researchers pushed for more precise specification of brain-behavior links, a looming concern is how to manage the estimation of a ever more complex linking function. In this talk, we discuss factor analytic linking functions as a strategy to conceptually group brain areas that function in a manner consistent with cognitive mechanisms assumed by overarching theories. From there, we discuss how one might apply constraints to factor analytic functions through LASSO regularization techniques as well using the structural information from the brain to constrain functional coactivation matrices.

Humans sample information from both external (stimulus-driven) and internal (memory-driven) environments to guide decisions shaped by objective (perceptual) or subjective (value-based) states of the world. Sequential sampling models, a prominent theoretical framework for examining the core computations underlying decision formation, propose that noisy evidence accumulates over time to a bound, with neurophysiological signals supporting these processes. However, little is known about the similarities and differences in the mechanisms driving decision-making across diverse types of decisions. In an experiment controlling for stimulus and motor differences, we apply joint modeling to analyze eye-movement and EEG recordings, examining how subjects sample stimulus-driven or memory-driven information to make perceptual and value-based choices.

Decision making and learning processes together enable adaptive goal-oriented behaviour. Animal studies demonstrated the importance of subcortical regions in these cognitive processes, but the human subcortical contributions remain poorly characterised. Here, we study choice and learning processes in the human subcortex, using a tailored ultra-high field 7 T fMRI imaging protocol combined with joint models. Joint models provide unbiased estimates of brain-behaviour relations by simultaneously including behavioural and neural data at the participant and group level. Results demonstrate relations between subcortical regions and the adjustment of decision urgency. Value-related BOLD differences were found with opposite BOLD polarity in different parts of the striatum. Multiple sub-cortical regions showed BOLD signatures of reward prediction error processing, but contrary to expectations, these did not include the dopaminergic midbrain. Combined, this study characterises the human subcortical contributions to choice and learning, and demonstrates the feasibility and value of joint modelling in facilitating our understanding of brain-behaviour relationships