CNS 246 Fall 1996

CNS 246 is primarily a research class. Students are expected to choose a project and make satisfactory progress on it by the end of the term.

Requirements

One or two people may work on a project. If you decide to collaborate we advise you to seek out someone with a different background to yours to work with. We would like to see engineers and biologists working together.

Short written project proposals will be due at the end of the third week of the term (see the class schedule). You may hand in hard copy of your proposal. Alternatively, you may put in on the web. We recommend using the latter approach, since then your proposal will be visible to the other students in the class. (If you are worried about making it available to the whole world, you may choose to restrict access to the page). We strongly encourage discussion and collaboration on projects.

If you choose to make a web page you are encouraged to keep it up-to-date as your project progresses. This will help make collaboration with the other students easier and also allow the TAs and instructor to find out how you're doing (and provide advice, if we can). Also, the complete web page can be turned in as your final project report.

If you do not use the web, you must turn in a hard copy report on your project by the last day of term (and if you use the web your final changes must be made by that day). We will be reviewing projects and assigning grades during finals week.

Data sets

We have provided a number of data sets collected by various labs at Caltech for use in this course. The files are available on this computer (mung) in the directory /data/cns246 (note: this link, which allows you to browse the directory and the README files, will only work if you are running your browser on mung). The data are not available on the web or through anonymous ftp.

You are free to use these data for your projects. However, please respect the generosity of the people who have collected and provided them. Do not give, or make available, any data to someone who is not taking the class (without explicit permission of the person who took them). If publishable work should result from your analysis of the data, it is your responsibility to settle the question of allocation of credit and authorship with the owners of the data before proceding.

Also, since the data sets are very large and space on mung limited, do not make verbatim copies of the data. You may place a symbolic link (see the ln(1) man page) from the central directory to your own if you wish. Obviously, if you filter or otherwise alter the data you will need to do so with your own copy (the data in /data/cns246 are write protected). If you make some alteration that is likely to be of general interest, please tell us and we may place a copy of the processed data in the central directory too, thus saving both disk space and needless duplication of effort.

Topic suggestions

For more information on one of these contact the person whose initials appear after the suggestion.

• Building a predictive burst model. Sorting algorithms must account for variability in the shape of extracellular spike waveforms from a single cell. A prominent source of such variability is bursting. Fee and Mitra have studied this variability empirically, however their analysis is based on the assumption that the underlying spike classification scheme is correct. You may instead, build a predictive model using a realistic channel model of an adapting soma (and possibly dendrite), along with boundary-element methods to predict the external field. (DK)

• Quantifying the performance of spike-sorting algrothims. A major hurdle in the development of automatic spike sorting turns out to be measuring the performance of the algorithm. Recorded neural data, of course, cannot be used since we don't know the correct classification. Synthesized data sets, on the other hand, must be formed making assumptions about the data, which are inevitably the same as the assumptions made by the algorithm being touted. You may either try to generate synthetic data from a sophisticated biophysical model with a range of assumptions. Alternatively, if it can be arranged, you may make simultaneous intra- and extra-cellular recordings. (JP or MS)

• Online (DSP) sorting implementation. Many of the algorithms suggested for automatic spike-sorting can in principle be implemented online, for example on a cluster of DSPs. We have such hardware available. You might study the feasibility of this approach, examine whether algorithmic changes would be necessary, or actually implement some online code. (MS)

• Independant Components Analysis for sorting. The spike sorting problem is similar in many ways to the blind signal separation problem studied in signal processing. Information maximization has proven to be effective in the latter case. You may try to apply this, or other blind signal separation approaches to the spike sorting problem. (JP)

• Non-Poisson firing and the JPSTH. The Joint Peri-Stimulus Time Histogram (JPSTH) was invented in 1969 to study the temporal relationship between spike trains recorded simultaneously. Later, Aertsen, Palm and others introduced the notion of "surprise" to provide a measure of signficance to the statistic. These measures implicitly require the individual trains to be Poisson (no self-correlation). What happens if they aren't? This project might involve both simulation using self-correlated neurons and theoretical calculation. (MS)

• Correlation through connectivity: the effect of the synapse. Aertsen and Gerstein have published a study of the effect of mutual connectivity on the cross-correlation of model cells. In this study they used a remarkably simple model. It would be interesting to investigate the effect of different synapses (AMPA, NMDA, GABAa etc.) on the expected cross-correlation using more realistic models. (CB)

• Structure in real data. This is a quite open-ended topic. You may use any of the techniques discussed in class, others that you may read about, or techniques of your own invention to explore the data sets provided (see above). (JP or MS)

---