Team members: Conor Keogh, Holly Wilson, Saptarshi Ghosh, Siobhan Hall
Animals update their behavioural strategies based on feedback to their decisions. Decoding strategies have demonstrated neural activity related to decision making. Similarly, feedback-related responses are well-established in electrophysiological data. Frameworks such as “predictive coding”, supported by evidence from sensory perception, have been proposed to unify prediction and feedback where an internal “model” of the external world is updated based on error. It remains unclear, however, how this feedback-related activity is integrated with predictive activity.
Demonstrating a relationship between predictive activity and reward activity on the previous trial would show that the brain “integrates” reward information to inform decision making. This would imply that some internal state is updated, providing electrophysiological support for models of decision making based on internal predictions. More specifically, we expect to find that the magnitude of neural responses vary as a function of correct vs. incorrect trials as a reflection of the error signal. We will consider a combination of regions of interest to investigate differential encoding of feedback through different functional areas.
Dataset (reference paper): Steinmetz
Determine whether feedback integration from a task (performed correctly with a reward, versus incorrectly with an auditory punishment) is evident in the subsequent task
Feedback integration from a task is used to update predictions in the subsequent trials (single trial history)
The methodology is further expanded within the code available at the following links:
- Load data
- Data exploration: Visualisation of the data
- Decode feedback and prediction information
- Determine optimal window for decoding
- Granger causality to compare reward[t] to response[t+1] where 't' corresponds to a relative trial
Reward: 'MEA' (cortical subplate), 'CA' (hippocampus), 'PT' (thalamus), 'MOp' (motor cortex), 'PAG' (midbrain), 'RT' (thalamus), 'COA' (motor cortex), 'BMA' (cortical subplate), 'CA2' (hippocampus), 'LSr' (basal ganglia)
Response: 'VISrl' (visual) , 'MG' (thalamus), 'NB' (midbrain) , 'CA2', (hippocampus) 'SPF' (thalamus), 'PIR' (motor cortex), 'CA' (hippocampus), 'EPd'(cortical subplate), 'POL'(thalamus), ' TT' (motor cortex)
Reward: 0.0 to 1.1 seconds (timelocked to reward at time 0)
Response: -0.8 to 0 seconds (timelocked to the response at 0)
The following links directly to the the 'CA' brain areas within the hippocampus - chosen as it is the area with the highest decoding accuracy for both response and reward.
To investigate whether the feedback-related activity "Granger-causes" the decision making activity on the next trial, we assess whether lagged values of the feedback window allow us to better predict future values of decision making activity than the previous value of decision making related activity alone.
Our results show there may be an association between reward and response in single trial history