Refine syllabus

[svenvanderburg]

Reproducible computational environments for research

• Virtual environments, docker, git

Neural data handling and preprocessing

• Spiking data
• LFP
• Calcium imaging
• Widefield imaging

Single cells analysis

• Place cells & grid cells
• Information methods and scoring methods
• Tuning curve construction
• GLM models of neural response

LFP

• Fourier/wavelet transform, filtering
• Spectrograms
• Cross – correlation
• Coherence
• Granger causality and phase slope index
• Phase-amplitude coupling
• Phase locking analysis
• Spike triggered average
• Spike-field coherence

Population-level analysis

• Neural decoding: linear classifier (SVM,logistic regression), Naïve Bayes, sequences
• Unsupervised learning: data segmentation and dimensionality reduction

[svenvanderburg]

Here are some of my thoughts about the syllabus: (we can discuss it further tomorrow)

Software engineering best practices for researchers

This could be given as a suggestion to teach before we dive into the 'real' content.

• Virtual environments. Can be a simple introduction to Conda, or more elaborate going through this material: https://carpentries-incubator.github.io/introduction-to-conda-for-data-scientists/
• Introduction to Git (Material could be: https://swcarpentry.github.io/git-novice/)
• Collaborative version control using git (At least 'centralized' workflow from https://coderefinery.github.io/git-collaborative/centralized/)
• Modular coding (https://coderefinery.github.io/modular-type-along/)
• Documentation (https://coderefinery.github.io/documentation/)
• Docker introduction (Material could be: https://carpentries-incubator.github.io/docker-introduction/) (In my opinion Docker is quite advanced, and not necessary for every scientist, I would first cover all above before diving into it)

Neural data handling and preprocessing

Instead of having a separate module for data handling and preprocessing I would suggest to have this interwoven in the other modules. I.e. you want to do some analysis, so you load in the data, preprocess it, then do your analysis. Unless this deviates a lot from what you already have of course?

• Spiking data
• LFP
• Calcium imaging
• Widefield imaging

Single cells analysis

• Place cells & grid cells
• Information methods and scoring methods
• Tuning curve construction
• GLM models of neural response

LFP

• Fourier/wavelet transform, filtering
• Spectrograms
• Cross – correlation
• Coherence
• Granger causality and phase slope index
• Phase-amplitude coupling
• Phase locking analysis
• Spike triggered average
• Spike-field coherence

Population-level analysis

• Neural decoding: linear classifier (SVM,logistic regression), Naïve Bayes, sequences
• Unsupervised learning: data segmentation and dimensionality reduction

It seems like now you focus a lot on teaching a broad spectrum of analysis techniques out there. In general I would focus mostly on giving students the right tools to be able to perform any preprocessing and analysis of any data they encounter in the wild. You can never cover all the techniques out there, but you can teach how to approach the problem in such a way that you can tackle it yourself. The modules could then be seen as 'examples' of a data analysis pipeline, at the end of the course students could develop such a pipeline themselves in groups. If you want to cover a lot of different analysis techniques you could give a lecture about it (without programming).

An example dataset-centered module, assuming a dataset with multiple single cells recorded from hippocampus during a spatial task:

1. Loading and exploring the data
2. Cleaning and preprocessing the data
3. Single cell analysis using PSTH
4. Place cells & grid cells
5. GLM models of neural response
6. Neural decoding of space

Maybe teach around a sort of 'neural data analysis workflow' with different steps (i.e. step 1: explore the data, step 2: don't reinvent the wheel, search for existing solutions, step n: share your work on github etc.)