Detection of Neuronal Assemblies by Frequent Item Set Mining
Filed under:
Electrophysiology
Sonja Gruen (Institute of Neuroscience and Medicine (INM-6), Forschungszentrum Jülich; Theoretical Systems Neurobiology, RWTH Aachen), David Picado-Muino (European Centre for Soft Computing, c/ Gonzalo Gutiérrez Quirós s/n, 33600 Mieres, Asturias), Denise Berger (Lab. Neuromotor Physiology, IRCCS Fondazione Santa Lucia, Via Ardeatina 306, 00179 Roma), George Gerstein (Dept. of Neuroscience, University of Pennsylvania, Philadelphia PA 19104), Christian Borgelt (European Centre for Soft Computing, c/ Gonzalo Gutiérrez Quirós s/n, 33600 Mieres, Asturias)
Hebb (1949) suggested cell assemblies as the building blocks of information processing in the brain. The member neurons are assumed to show correlated activity. We present a data mining method that detects assemblies in massively parallel spike data both reliably and efficiently.
Gerstein et al. [1] developed an accretion approach to detect joint spiking patterns in parallel spike trains. Starting from single neurons, this approach iteratively accretes neurons into sequences as long as another neuron shows significantly correlated activity with the accreted neurons. However, accretion suffers from several drawbacks: it works on sequences instead of sets, thus incurring high costs from redundant detections (memory consumption, speed) and may also miss assemblies.
Here we present an alternative approach based on frequent item set mining (FIM) that amends these drawbacks and was developed for finding sets of items (here: neurons) that frequently occur (here: spike) together. FIM algorithms efficiently count joint spiking events that exceed a given minimum support (occurrence frequency) by efficiently exploring the complete search space without redundancy. The found patterns may be assessed statistically by taking the maximum p-value over all one-neuron-against-rest tests and by additional subset conditions. We examined (a) no subset conditions, (b) weak subset conditions (existence of a stepwise significant sequence as in accretion), and (c) strong subset conditions (all possible sequences must be stepwise significant).
The false positive (FP) and false negative (FN) rates are evaluated under different subset conditions. Interestingly we found that FIM without any subset requirements and without any statistical test leads to the same or even better results as accretion or FIM with weak subset conditions. FIM with strong subset conditions reduces FPs but at the price of a considerable increase of FNs. This leaves us with a fast and reliable plain FIM algorithm (finding maximal frequent item sets), enabling a conclusive statistical test based on surrogate data (cf. [2]). Our next steps will be to test the method in cases where more than one assembly is present and to explore dynamic assembly processing.
Acknowledgements: Helmholtz Alliance Systems Biology, EU (FP7-ICT-2009-6, BrainScaleS)
References
[1] Gerstein et al (1978) Brain Res 140: 43-62
[2] Gerstein et al (2012) J Neurosci Methods 206: 54-64
Gerstein et al. [1] developed an accretion approach to detect joint spiking patterns in parallel spike trains. Starting from single neurons, this approach iteratively accretes neurons into sequences as long as another neuron shows significantly correlated activity with the accreted neurons. However, accretion suffers from several drawbacks: it works on sequences instead of sets, thus incurring high costs from redundant detections (memory consumption, speed) and may also miss assemblies.
Here we present an alternative approach based on frequent item set mining (FIM) that amends these drawbacks and was developed for finding sets of items (here: neurons) that frequently occur (here: spike) together. FIM algorithms efficiently count joint spiking events that exceed a given minimum support (occurrence frequency) by efficiently exploring the complete search space without redundancy. The found patterns may be assessed statistically by taking the maximum p-value over all one-neuron-against-rest tests and by additional subset conditions. We examined (a) no subset conditions, (b) weak subset conditions (existence of a stepwise significant sequence as in accretion), and (c) strong subset conditions (all possible sequences must be stepwise significant).
The false positive (FP) and false negative (FN) rates are evaluated under different subset conditions. Interestingly we found that FIM without any subset requirements and without any statistical test leads to the same or even better results as accretion or FIM with weak subset conditions. FIM with strong subset conditions reduces FPs but at the price of a considerable increase of FNs. This leaves us with a fast and reliable plain FIM algorithm (finding maximal frequent item sets), enabling a conclusive statistical test based on surrogate data (cf. [2]). Our next steps will be to test the method in cases where more than one assembly is present and to explore dynamic assembly processing.
Acknowledgements: Helmholtz Alliance Systems Biology, EU (FP7-ICT-2009-6, BrainScaleS)
References
[1] Gerstein et al (1978) Brain Res 140: 43-62
[2] Gerstein et al (2012) J Neurosci Methods 206: 54-64
Preferred presentation format:
Poster
Topic:
Electrophysiology
Latest news for Neuroinformatics 2011
Follow INCF on Twitter
