Final Project

The goal of the final project is for you to explore a few of the topics in the course in a bit of depth. You need to choose one of the three sub-assignments below. You may choose to do additional problems as extra credit. Problems 1 and 2 are essay problems (~1000 words), and problem 3 involves coding. In addition to the references given, it is expected that you will cite at least 3 other papers. Make sure to cite your sources and include a reference section/bibliography at the end. Feel free to include figures from sources that you cite or create your own. You are welcome to propose an alternative topic/problem – please do so before the end of classes.

  1. Technologies for measuring the activity of all the neurons in the brain.

    In class, we discussed a variety of techniques to measure the activity of neurons in the brain. Imagine that you’ve been given a nearly infinite budget by a philanthropist to embark on a ten year program to try to simultaneously record the activity of as many neurons in the brain of a mouse as possible. Would you fund electrode development, microscopy techniques, other approaches or all of the above? What are the benefits and limitations of each approach? What do you think is currently the biggest gap in our technologies in terms of measuring a specific part or fraction of the brain?

    References:

    1. https://www.nature.com/articles/s41592-021-01257-6

    2. https://iopscience.iop.org/article/10.1088/2515-7647/ac76f9/meta

    3. https://www.nature.com/articles/s41551-022-00941-y

    4. https://www.biorxiv.org/content/10.1101/2021.10.13.463862v1.abstract

    5. https://www.science.org/doi/full/10.1126/science.abf4588

  2. Artificial vision: the challenge of precisely controlling neural circuits

    Imagine that you want to create a brain implant that will produce artificial vision. Specifically, assume that your goal is to precisely control the activity of individual neurons in primary visual cortex. Compare optogenetic solutions with solutions based on electrodes. Questions that you likely want to address: What are the benefits and drawbacks of each approach in terms of:

    • controlling neural activity precisely,

    • targeting the right neurons,

    • logistically placing/controlling the interface (i.e., surgeries, implants, system size/weight),

    • ethical concerns?

    • other challenges

    References:

    1. https://www.nature.com/articles/s41593-021-00902-9

    2. https://www.nature.com/articles/s41593-017-0018-8

    3. https://www.sciencedirect.com/science/article/pii/S0896627320308072

    4. https://iopscience.iop.org/article/10.1088/1741-2552/abd0ce/meta

    5. https://www.science.org/doi/full/10.1126/science.abd7435

    6. https://www.sciencedirect.com/science/article/pii/S0092867420304967

  3. Decoding place cell activity

    In class, we used Bayes rule to decode the choices of reaches. In this problem, you will use Bayes rule to decode the position of a rat in a maze.

    In the references below, you’ll see some example code that calculated place fields and decoded the position of a rat in 2D. For the final project, you need to do something similar. The data for the final project is found in the file final_project_data.nel. This comprises the position data and spiking activity for a rat running in a linear track. Thus the position is 1-dimensional rather than 2-dimensional (as in the tutorial). Example code showing how to load it is found in final_project_read_data.ipynb Half way through the session, the maze was shortened, as described in this paper. Your task for the final project is to explore decoding this data. At minimum, you need to do the following:

    1. Following the tutorial in reference 2, use place cell activity to decode the animals position as they run through the maze. Note that the ratemap will be 1-dimensional (rate vs space) for this project rather than 2-dimensional, so if you want to calculate it using the helper functions, you’ll need to use TuningCurve1D rather than TuningCurve2D. Run the whole process only on the first, long-track portion of the data.

    2. Modify the code so that you use the first K laps to estimate the ratemap rather than the entire long-track period. Calculate the decoding error for the subsequent laps. How does changing the value of K change the error? (Note that if you are following the tutorial, when you calculate the run epochs, laps will correspond to pairs of run epochs - one each for each direction on the track.) Plot the actual and decoded trajectories.

    3. Use the place field ratemap for the long track to decode the position on the short track. Compare this with what you get if you warp the position on the short track before calculating the error. If the position data is in a short_track_pos nelpy object, you can warp it by doing:

      warped_short_track_pos = short_track_pos
warped_short_track_pos.data = (warped_short_track_pos.data - offset) * scaling_factor

      where you’ve figured out by eye or by calculation what the offset and scaling_factor should be for the short track to span the same length as the long track.

    4. Optionally you can do more analyses (the effect of the number of neurons, the effect of different smoothing factors on decoding performance like in this paper https://onlinelibrary.wiley.com/doi/abs/10.1002/hipo.22714, etc.). If you are interested in doing replay analysis, please reach out and we can share the times of SWR with you.

    References:

    1. https://journals.physiology.org/doi/full/10.1152/jn.1998.79.2.1017

    2. https://notebook.community/Summer-MIND/mind_2017/Tutorials/SpikeDecoding/spike_decoding_python