Miniproject for Oxford Interdisciplinary Biosciences DTP data management and statistics course. Example data from 2-photon calcium imaging of mouse somatosensory cortex.
- Download git bash for windows https://gitforwindows.org/
- make a code directory (e.g. Documents/code)
- use the git bashj to clone the d2p type: 'git clone https://github.com/sarmstg/d2p.git' into git bash
- download anaconda https://www.anaconda.com/distribution/
- run jupyter notebook from anaconda navigator
- Create a new notebook in which you can work
Tasks to get started
- Load the .npy data from the repo data folder
- Make some plots of the fluoresence data on different trial types and outcomes, do you want to average at all?
- Is there more neural activity on go and nogo trials? How about hit or miss trials? (Does this depend on how you average??
- Push your new notebook back to the repo in the jupyter folder
What is the structure of the data?
- The session1.npy file is a 3d array of shape [n_cells x n_trials x time]
- The trial_info.npy dictionary has values that tell you about the trial (each array should have the length n_trials)
Potential project focus?
- Can the trial outcome (hit vs miss) be predicted from activty preceeding stimulation? This would involve analysing if there is a difference in the first 5 frames [0,1,2,3,4] of hit and miss trials.
How can you tell if there is a difference in neural activity?
- The first pass analysis would be to take a mean across the first 5 frames, so you have a matrix of shape [n_cells x n_trials]. This matrix can be passed to several differnet methods of classification e.g. a Support Vector Machine or a logitic regression (you will need to split your data into test and training). If this doesn't work, we can come up with an alternative to averaging across frames.
The two presentations shown at the start