Kevin Bender, Professor

Closed (1) Fitting neuronal models to electrophysiological data.

Check back for status

Studying the biophysical properties of neurons, such as the different ion channel distributions and kinetics, can be done by fitting electrophysiological recordings from acute slices to neuronal compartmental models. In this process, a detailed neuron is modeled to a set of electrical circuits that describe its biophysical properties. These properties, also called model parameters, can be adjusted. We try to constrain thein silico model’s parameters to behave like in vitro neurons recorded during live experimentation. This process is done using an optimization algorithm that relies on training the model across thousands of permutations with respect to recorded physical stimulations from an actual neuron. We are investigating novel approaches to perform such optimization. A further goal of this project is to adapt the optimization so that it can work on models that differ in physiology and research goals.

We are seeking undergraduate students who are interested in computational neuroscience and would like to take part in this project. While the project requires computer science skills often taught to EECS/CS students, being one of these majors is not required. Possible tasks include:
1. Using and modifying an optimization algorithms library written in Python language.
2. Running computations on Linux Clusters or GPUs, as well as managing high-performance computing libraries and dependencies for these computations

3. Adapting and creating models in the NEURON language for neuronal cell modeling.
4. Standardizing and deploying this software to run across multiple native environments.

We strongly encourage students from underrepresented minorities (URM) to apply, and we might be able to offer NIH funding to URM students that can commit to long-term research projects.


Day-to-day supervisor for this project: Roy Ben-Shalom, Post-Doc

Qualifications: Knowledge about Python language or any Object-oriented language is required. . Some knowledge about neurons or electrical circuit theory will be helpful but isn’t essential, though an interest in neuroscience is. Some knowledge about linear algebra or optimization theory is required. Applicants should mention if they have any experience with data pipelines, hdf5, shell scripting, technical documentation, or GUI design.

Weekly Hours: 9-11 hrs

Off-Campus Research Site: Mostly remotely.

Related website: https://benderlab.ucsf.edu/lab-members

Closed (2) Using Deep Neural Networks to identify parameters in neuronal models

Check back for status

The goal of this project in general is to predict the biophysical properties of a neuron based on its electrophysiological response to stimuli. We made significant progress in using one CNN trained to predict the properties of one neuronal subtype. Next we are trying to build one CNN to predict any neuronal subtype. Students in this project will have the opportunity to interact with various stages of the machine learning process, from data generation and analysis to neural network training. And general data analysis




Most of the DL part of this project is already done! We are now mostly in the stages of generating more data and training CNNs. Another big task is analyzing the data set in different ways (PCA RF and more)

We are looking for students with an interest in machine learning, optimization, statistics and/or neuroscience with good programming skills to help us with these tasks.

We strongly encourage students from underrepresented minorities (URM) to apply, and we might be able to offer NIH funding to URM students that can commit for long term research projects.

Qualifications: In particular, CS 61A and 61B would be ideal. Machine learning knowledge is not required, though any previous experiences in related areas would be useful.

Weekly Hours: 9-11 hrs

Off-Campus Research Site: 2119 5th St

Related website: https://www.biorxiv.org/content/10.1101/727974v1

Closed (3) Accelerating neuronal simulation on using GPUs

Check back for status

We have developed a framework that runs neuronal biophysical simulation on GPUs. And accelerates their runtime by two orders of magnitude. for more details please see https://www.biorxiv.org/content/10.1101/727560v1?rss=1


Developing new features for NeuroGPU, which will primarily involve writing highly optimized scoring functions on GPUs using CUDA/C++. As such, experience with C++ compiling and CUDA is an advantage. Experience with python is required.

We strongly encourage students from underrepresented minorities (URM) to apply, and we might be able to offer NIH funding to URM students that can commit for long term research projects.



Qualifications: Python required CUDA is an advantage

Weekly Hours: 9-11 hrs

Off-Campus Research Site: 2119 5th St

Related website: https://github.com/roybens/NeuroGPU

Closed (4) Simulating variants of Neurodevelopmental diseases

Check back for status

This is very similar to the Project #1 but requires some neuroscience understanding. Here we would fit ion channel models to experimental recordings from mutated variants of the channel associated with neurodevelopmental diseases such as Epilepsy and Autism

Developing and running our fitting procedure
A neuroscience literature search in regards to channelopathies and anti-epileptic drugs

Qualifications: Strong python programming skils Basic neuroscience is an advantage

Weekly Hours: 9-11 hrs

Off-Campus Research Site: 2119 5th St