CorbanSwain / 9.523-Prediction-of-Functional-Connectomes

Proposal Requirements

9.523 | Fall 2018

The project proposal should be submitted through Stellar. The deadline is on Nov. 9th, but feel free to submit it as soon as you have it. You can get feedback and get started with the project earlier if you want.

Tentative Timeline

• Week 0, Nov 5 - 9: Prepare proposal and write out pseudocode.
• Week 1, Nov 12 - 16: Background, identify what's already been done and
  find existing tools that we can use.
• Week 1 & 2, Nov 12 - 23: Set up the framework for the implementation and generate first version of the project code.
• Week 2 & 3, Nov 19 - 30: Iterate on the implementation based on preliminary results.

The project proposal should contain:

• Title: Recapitulation of Functional Connectomes from a Simulated Cortical Hypercolumn

• Team members:

  • Corban Swain
  • Nick Ning

• Abstract:

  1. One or two sentences providing a basic introduction to the field.

     • Physiological recording of brain activity in organisms from C. elegans to Homo sapiens is one of the primary methods for furthering our understanding of the brain.
     • Part of this "understanding" lies in the ability to elucidate the topology of effective connections between neurons--i.e. building brain networks (Bullmore, 2009).
     • A better understanding of brain networks can be applied to build
       more realistic models neuronal computations and generate more physiologically-motivated framework for diseases of the brain.

  2. Two to three sentences of more detailed background and previous work.

     • Existing techniques for recording brain activity span a range of temporal and spatial resolving abilities (see Table 1 for a limited summary of these methods).
     • In the context of this proposal, we give attention to opportunities provided by recent developments in genetically-expressed fluorescent ion and voltage reporters: a technology which, in combination with volumetric optical imaging, can be used to record from individual neurons across large regions of the brain in transparent organisms like zebrafish.
     • Existing techniques for generating a network graphs from brain
       recordings are built for recordings from electrodes and fMRI and rely primarily on correlative techniques and produce undirected connectivity models.

  3. One sentence clearly stating the general problem being addressed.

     • In this proposal we will begin to address gap in existing techniques to generate directional graphs of neuron connections from functional recordings of brain activity and to effectively harness the power of large volume, near single-cell recordings made possible by fluorescent reporters with optical imaging.

  4. One sentence summarizing the expected main result (enumerate multiple possibilities).

     • We expect to demonstrate in silico that a recurrent neural network
       (RNN) can be trained using activity recordings from many small known networks to accurately predict the directional graph of an unobserved network. We might find that RNNs are also capable of predicting the input to a simulated unobserved network. Additionally, we have considered the possibility that different machine learning architectures would each be more accurate at predicting different aspects of the network topology.

  5. Two or three sentences explaining what the main result is expected to reveal in direct comparison to previous work.

     • The successful completion of this work would provide a novel
       computational tool for the analysis of functional brain recordings.
     • Specifically, we expect our proposed tool to be able to to use fluorescence-based recordings (after image preprocessing) to infer functional connectomes with directional information ((need justification)). We also expect that our tool will be well posed, unlike correlative methods, to integrate the prediction of the synaptic properties (e.g. reeptor type ((i need better exaples here))) of neuron connections because of its ability to train on recordings annotated with information beyond its functional connections.

• Lecture/s related to the project: explain how the project relates to one or more lectures from 9.523 (two sentences per related lecture)

  • lec 3 - Hippocampal mechanisms of memory and cognition
  • lec 7 - Functional Modules: what good are they and how do we get them?
  • lec 9 - (something goes here)

• Methods: Plan of the experiment/s to execute during the project. Detail computational resources needed or any material needed for the psychophysics experiments.

  1. Model of Cortial Hypercolumn [[Hansel 1998]]

Brainstorming

• Original Pitch - I am interested in working on simulation of temporal and phase encoding for learning and information transmission within a learning network. Very open to other ideas as well.
• The Goal - The ability to predict network connectivity from individual neuron recordings.
• The Method - To use the simulation of a simple network to
• Neuron Recordings - The use of probes to detect and record the activity of neuronal cells over time. There are many technologies which can be used to detect the activity of neurons.
  • Patch clamps can directly measure the difference in electrical potential (i.e. voltage) between the inside and out side of a neuronal cell.
  • Electrodes can measure the electrical potential fields emanating from current-conducting neurons.
  • Genetically expressed fluorescent reporters (e.g.)
• Have a logical flow to the paper ... address a somewhat relevant.
• We propose applying machine learning to both predict the state and connections of a small, simulated network of spiking neurons.

Discard Pile

• There exist many methods to record activity from the brain of
• Importance of constructing network connectivity from activity data/recordings.
• Explanation of old methods for inferring connectivity from activity time courses.