DataSciClubCUMC

Author: Katherine Shakman Description: My projects related to the CUMC Data Science Club.

Project proposals for the Hackathon (Dec 2 2017):

Project 1: For a pair of neurons a and b, determine whether activating neuron a will increase the correlation between the activity patterns of the two neurons a and b.

• Language: Python 3.
• Data: My own.

1. For a pair of example traces, calculate their correlation coefficient (use of Spearman coefficient seems appropriate for my data per: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3576830/).
   * Load the traces (saved as .mat files) into python. (See http://www.blogforbrains.com/blog/2014/9/6/loading-matlab-mat- data-in-python)
2. For n pairs of traces with condition 1 (no activation of a) and m pairs of traces with condition 2 (activation of a), find the cross-correlations between each pair and compare correlations in condition 1 with correlations in condition 2 (e.g. are the corrleations higher in one condition than the other?) * Calculate the correlation between each pair of recordings.
3. Get summary statistics and visualize. * Get the median, mean and SD of the correlation in condition 1 and in condition 2.
   * Plot the distributions of correlations.

Project 2: Train an SVM to decode stimulus type from neural activity.

• Language: Python 3.
• Data: My own.

1. Generate labels of the stimulus type for each traces.
2. Split the available traces into a training set, a crossval set, and a test set.
3. Train an SVM sklearn's SVM package.
4. Test on the CV set.
5. Once satisfied that everything is working, try the test set.
6. Visualize the results.

Future Project Idea: Implement a random forest and try using it to categorize my neural recordings by stimulus type.