Generalised linear modeling

EN.580.439 Models of the Neuron

Model details

• Constructed 3 GLMs after initial classification of cells into 3 groups based on their spatial and temporal firing patterns
• A data-driven approach was used in determination of each model's covariates
• All GLMs have all positional covariates
• Each GLM has their own velocity, speed and direction covariates
• Each GLM has their own history dependence

Model 1

Model 2

Model 3

Results

Conclusions

• Model 3 performed slightly better when comparing relative goodness-of-fit due to low number of covariates
• However, all three models are very similar
• Positional covariates had the most explanatory power
• All the neurons showed some history dependence between 5-20 ms in the past as well as around 90-140 ms