Computer Game, Qualitative, and MEG/EEG Assessment of Serotonergic Psychedelics

Hierarchial Gaussian Filter (HGF)-estimated average precision-weighted prediction error change in the conditioned hallucinations and probabilistic reversal learning tasks

Participants' trial-by-trial responses on auditory and visual conditioned hallucination (CH) tasks as well as a probabilistic reversal learning (PRL) task are fitted with an enhanced Hierarchial Gaussian Filter (eHGF) model of hierarchial belief updating under perceptual/informational uncertainty using eHGF package in Julia. Precision weighted prediction errors used to update the 2nd and 3rd levels are estimated for each trial for each individual. Average precision weighted prediction errors across all trials as well as trials with greatest changes (0 stimulus trials for CH tasks and the first 5 trials after reversal in PRL) will be calculated. Change relative to baseline will be calculated. Greater precision weighted prediction errors reflect a greater weighting of new compared to previously learned information and greater belief updating.

Auditory and visual conditioned hallucination rate change

Participants indicate detection of an auditory tone or visual pattern within auditory or visual noise. These stimuli are paired with cross-modality cues, causing participants to subconsciously expect to detect the stimuli using pavlovian conditioning. This conditioning causes participants experience "conditioned hallucinations" where they perceive the stimuli when it is not present. Conditioned hallucination rate is calculated as number of trials in which no stimulus was present but participants indicated perceiving the stimulus divided by the total number of trials in which no stimulus was present. A greater conditioned hallucination rate indicates greater reliance on previously learned information over incoming sensory evidence and closely tracks state-level hallucination susceptibility.

Perseverative and jumping-to-conclusion error total change in probabilistic reversal learning task

Participants choose one of three images over multiple trials and receive reward feedback (low vs. high) in the PRL task. Each image has different reward probabilities that are unknown to participants. Through trial and error, participants learn which image is most rewarding and aim to maximize winnings. Over time, reward probabilities reverse without warning, requiring participants to update their beliefs. Every trial is scored as a perseverative error, a jumping to conclusions error, or neither, and then total number of error types are counted. Perseverative error trials occur when participants select a high reward probability image after its reward probability switches and they receive negative feedback. Jumping to conclusions (JTC) errors occur when participants switch from the highest reward stimulus to a different stimulus when there has been no reversal.

Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q) score change

The Quality of Life Enjoyment and Satisfaction Questionnaire - Short Form (Q-LES-Q-SF) is a 16-item self-report questionnaire that captures life satisfaction over the past week. Each question is scored on a 5-point scale from 1 (Very Poor) to 5 (Very Good) that indicates the degree of enjoyment or satisfaction achieved during the past week relative to the particular activity or feeling described in the item. Scores range from 16 to 80. Higher score indicates higher quality of life.

Patient Health Questionnaire-9 (PHQ-9) score change

The Patient Health Questionnaire-9 (PHQ-9) consists of 9 items, each scored on a four-point scale from 0 = "Not at all" to 3 = "Nearly every day." It measures the severity of depressive symptoms over the past two weeks. The total score is calculated by summing all 9 items, with higher scores indicating more severe depression. Scores range from 0 to 27 and are typically interpreted as follows: 0-4 = minimal depression, 5-9 = mild, 10-14 = moderate, 15-19 = moderately severe, and 20-27 = severe depression.

Change in learning rate difference between noisy and volatile learning conditions in a probabilistic learning game

Participants complete a probabilistic learning game in which they estimate where objects will fall from an unseen spaceship, receiving trial-by-trial feedback on the actual location of the falling object. The location of the falling objects are sampled from a series of normal distributions that vary by mean and standard deviation such that changing means corresponds to high environmental volatility and high standard deviation correspond to high noise. Every trial produces a prediction error between the participant guess of where the object will fall and the actual position. Learning rate (LR) is calculated on each trial by dividing the subsequent change in participant estimate by the difference in estimate and actual object location (the prediction error). Average LR for each of 5 levels will be calculated. Higher average LRs reflect greater weighting of prediction errors and learning from new information under each level's condition.

Change in jumping to conclusion and perseverative errors in a feedback-free probabilistic learning game

Participants must guess which of 3 ponds a colored fish was drawn from. They can see the proportion of each color of fish in each pond. Without warning, the pond that fish are drawn from are switched. Trials in which participants select the pond from which fish were previously drawn after a new pond is being sampled from are perseverative errors. Trials in which a participant mistakenly believes that a selection of a rare fish from the pond represents a shift to a new pond are marked as jumping-to-conclusion (JTC) errors. Total perseverative and JTC errors are calculated for each participant. More perseverative errors indicates difficulty updating beliefs and overreliance on prior compared to new information. More JTC errors indicate oversensitivity to new information compared to previously learned information as well as difficulty distinguishing probabilistic noise from environmental volatility.

Reward and punishment learning rate changes in probabilistic reversal learning task

Participants' behavior in the PRL task will also be modeled using Rescorla-Wagner reinforcement learning as a function of weighted updates from predictions errors. A single learning rate parameter - the weighting of prediction errors - is estimated for each participant. The model will also be run calculating two learning rates for each participant - differing by whether the prediction error was positive vs. negative (positive vs. negative outcome). Higher learning rates reflect greater reliance on new as opposed to previously learned information.