This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. We wish to utilize the computing facilities to analyze the responses of neurons in the central auditory system. Specifically, we wish to compute the response functions of auditory neurons using a novel computational methodology. Under a previous collaboration with Tatyana Sharpee, formerly of UCSF and now at the Salk Institute, the supercomputer facilities were utilized to perform a similar analysis, though now we would like to continue the methodology but on a different set of data from subcortical and cortical stations. The computational methodology we will implement has been previously published (Atencio et al., 2008). We will compute the receptive fields of auditory neurons. The receptive field describes the relationship between the stimulus and the response of a neuron. The receptive field can be approximated as a set of linear filters, each of which may be calculated by maximizing the information between the stimulus and the neural spiking response. Thus, each filter is termed a maximally informative dimension (MID). Briefly, the first MID is the direction, or dimension, in stimulus space that accounts for the most mutual information between the stimulus and the response. We obtain the first MID through an iterative procedure, where the relevance of any """"""""candidate"""""""" dimension V is quantified by computing the mutual information between the occurrence of single spikes and projections of the stimulus onto V. We search through different directions in the stimulus space until convergence. Upon finding the first MID, we then estimate a second MID. The second MID is the dimension in the stimulus space that, together with the first MID, further maximizes the information. The stimulus is approximately 15,000 different stimulus spectrograms, each having 500 pixels. A direction in stimulus space is a spectrogram, where the pixels in the spectrogram may take on any value. The computational algorithm searches through the stimulus spectrograms till it converges to a single direction, or image, which is the MID. Since each pixel in a spectrogram may take on multiple values, searching and converging to the appropriate pixel values over this data set is computationally intensive, and thus ideally suited for implementation on the supercomputer facilities. From previous work with Dr. Sharpee, where she used the supercomputer facilities, we know that to calculate two MIDs for one neuron takes approximately 160 hours. Thus, given our data set size, we wish to request 75,000 hours of facility service units, as well as 2 terabyte of disk space. The work from our collaboration with Dr. Sharpee has already led to publication in an high impact journal, and we anticipate that these further computations will be similarly fruitful. Atencio CA, Sharpee T, Schreiner CE (2008) Cooperative nonlinearities in auditory cortical neurons. Neuron 58:956-966.
Showing the most recent 10 out of 292 publications
