Model-Based Planning Deficits in Compulsivity Are Linked to Faulty Neural Representations of Task Structure
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Article activity feed
-
-
###Reviewer #2:
In this work the authors seek to disentangle the reason for a well-documented effect regrading reduced model-based tendencies among high compulsive individuals. The authors collected behavioral and EEG data from ~200 participants performing a two-stage decision task. Main findings show a latent compulsivity factor is associated with weaker transition-type effects at the task's 2nd stage. Specifically, high compulsive individuals show smaller reaction-times and parietal-occipital alpha-band power differences between uncommon and common transitions. These findings are interpreted as evidence in favor of less accurate model as a reason for reduced deployment of model-based strategies in compulsive individuals. Authors further note reduced theta power for compulsive individuals during 1st stage choice.
I am generally very …
###Reviewer #2:
In this work the authors seek to disentangle the reason for a well-documented effect regrading reduced model-based tendencies among high compulsive individuals. The authors collected behavioral and EEG data from ~200 participants performing a two-stage decision task. Main findings show a latent compulsivity factor is associated with weaker transition-type effects at the task's 2nd stage. Specifically, high compulsive individuals show smaller reaction-times and parietal-occipital alpha-band power differences between uncommon and common transitions. These findings are interpreted as evidence in favor of less accurate model as a reason for reduced deployment of model-based strategies in compulsive individuals. Authors further note reduced theta power for compulsive individuals during 1st stage choice.
I am generally very impressed with this manuscript. I think the authors are addressing an important question that has a lot of promise in pushing the field forward. I also believe that given the number of participants, this is a relatively well powered EEG and behavioral study. Yet, I have one major concern as detailed below:
The authors relay on 2nd stage effects to estimate to what extent individuals are more or less aware of the transition structure (i.e., to what extent individuals are surprised by an uncommon state, or unsurprised by the common one). However, unlike Konovalov & Krajbich, 2020 who used a mouse tracking procedure to capture participants' 2nd stage expectation, both the RT and alpha band scores might be confounded due to 1st stage choice strategy. Individuals with stronger deployment of model-based strategies in the 1st stage tend to get more often to the best 2nd stage choice by means of a common transition. In contrast, the choices made by MF individuals at the 1st stage will not direct them more often to the best 2nd stage option by means of a common transition. This means that for a MF individual, the overall value difference for the two options offered at the 2nd stage will be similar in common and rare transitions, while for a MB individual the value difference will be higher in common vs. rare transitions. This is even when both MB and MF agents have a perfect knowledge regarding the task transition structure, and are equally surprised by an uncommon transition. Since the 2nd stage decision is easier on average on common vs. rare transitions for MB agents, they should also exert stronger transition effects compared with MF agents on 2nd stage estimates. One such effect might be greater alpha-band on rare transitions reflecting a greater mental effort (as the authors note). Also, when the decision is easier due to larger value difference, shorter RTs are to be expected (e.g., Pedersen et al., 2017 on pbr; Shahar et al., 2019 on plos-cb). This means that transition effect on both alpha-band and RTs is expected due to the use of MB strategies in the 1st stage, even if transition probability is perfect. Indeed, the authors report lower MB deployment at the 1st stage for compulsive individuals, which is in-line with their weaker transition-related effects on the 2nd stage.
-
###Reviewer #1:
In this report, the authors test a hypothesis about the nature of high-level ("model-based") vs. low-level ("model-free") learning across the spectrum of behavioral compulsivity. Prior literature has suggested that high-compulsive individuals have a deficit in either forming a model of the world, or implementing that model due to competition from learned low-level action-outcome tendencies. This report tested a large number of participants with concurrent EEG (N=192) across a range of compulsivity on the well-known two-step reinforcement learning task.
The authors note that they "replicated prior work in findings that individual differences in compulsivity and intrusive thought ... were associated with reduced model-based planning" with analyses of accuracy (pg. 8). Analysis of RT revealed a novel effect of compulsivity …
###Reviewer #1:
In this report, the authors test a hypothesis about the nature of high-level ("model-based") vs. low-level ("model-free") learning across the spectrum of behavioral compulsivity. Prior literature has suggested that high-compulsive individuals have a deficit in either forming a model of the world, or implementing that model due to competition from learned low-level action-outcome tendencies. This report tested a large number of participants with concurrent EEG (N=192) across a range of compulsivity on the well-known two-step reinforcement learning task.
The authors note that they "replicated prior work in findings that individual differences in compulsivity and intrusive thought ... were associated with reduced model-based planning" with analyses of accuracy (pg. 8). Analysis of RT revealed a novel effect of compulsivity on model-based planning, which was replicated using archival data of the same task. E-phys findings indicated that the candidate biomarkers of control in P300 and frontal midline theta were unrelated or not specifically related to model-based planning deficits in compulsivity, respectively (more on this below). However, the novel biomarker of posterior alpha power during the transition period was indeed linked with model-based planning deficits in compulsivity. This is novel.
This report is extremely well motivated by prior literature, it is very well written, and very well executed. Supplemental controls for age and IQ, tests of the specificity of EEG effects with compulsivity, and tests of the specificity of this compulsivity dimension on dependent measures in relation to associated personality variables (e.g. anxious depression & social withdrawal, also raw item measures) all work together to bolster the conclusions. This is a very carefully presented report.
Despite these virtues and advantages, the take-home message that I leave with is that EEG is not ideally suited for revealing the nature of compulsivity on model-based planning. P300 was irrelevant, frontal theta was possibly indirectly related (see below), and only posterior alpha was indicative of the compulsivity-related findings revealed in the behavioral analysis. This is unfortunately the least useful assessment of cognition used here, as it reflects the lowest level of control or decision making amongst these EEG measures. This perceptual effect is likely more of a consequence of the behavior than a candidate mechanism underlying it. This conclusion unfortunately diminishes the utility of these findings.
Regarding theta: Theta power and compulsivity were related to RT change, and they were related to each other, even though theta was not related to model-based planning (presumably tested via accuracy / choice). Although these patterns are carefully interpreted, it isn't perfectly clear how these were tested and I suspect there may be more that could be tested / inferred here. First, theta may still be related to the latent feature of "model-based choice" even if it is not significant due to the manifest measure based on choice patterns. This requires some careful unpacking of semantics and what latent constructs can be inferred from which manifest variables, but it is always a good idea to question what a single measure can infer about complex cognitive states. Second, taking this theoretical issue and including a methodological point, the single trial theta-RT relationship may still be altered by compulsivity even if theta power is not. Power and power-RT correlations have been presented as different measures of control that can be differently affected by a host of variables. This could presumably be tested by a thetaRTcompulsivity interaction, and could be visualized as a correlation between the individual theta*RT beta weight (Y-axis) with compulsivity on the X-axis.
-
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 1 of the manuscript.
###Summary:
The authors aimed to disentangle the processes underlying compulsive individuals' difficulties forming models of the world, or implementing such models due to competition from lower-level action-outcome tendencies. To this end, they obtained behavioral and EEG data from ~200 participants performing a well-established two-step reinforcement learning task. The authors note that they "replicated prior work in findings that individual differences in compulsivity and intrusive thought ... were associated with …
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 1 of the manuscript.
###Summary:
The authors aimed to disentangle the processes underlying compulsive individuals' difficulties forming models of the world, or implementing such models due to competition from lower-level action-outcome tendencies. To this end, they obtained behavioral and EEG data from ~200 participants performing a well-established two-step reinforcement learning task. The authors note that they "replicated prior work in findings that individual differences in compulsivity and intrusive thought ... were associated with reduced model-based planning" with analyses of accuracy. RT analyses revealed a novel effect of compulsivity on model-based planning, which was replicated using archival data of the same task. EEG findings indicated that the P300 and frontal midline theta, both well-established measures of cognitive control, were unrelated or not specifically related to model-based planning deficits in compulsivity, respectively. Posterior alpha power during the transition period, a novel marker, was linked with model-based planning deficits in compulsivity.
-
