Receptive field characterization by spike-triggered independent component analysis

The spikes generated by a neuron in response to stimuli provide information about the nature of the stimuli and also about the functional organization of the circuit in which the neuron is embedded. Spike-triggered analysis techniques such as spike-triggered covariance (STC) have been proposed to characterize the receptive field properties of neurons. So far, they have been able to provide only limited information about the functional organization of neural circuitry; in particular, STC tends to generate subfields that are mixed observations of independent processes. We address this problem by adding a criterion that sources are independent, resulting in an approach we call spike-triggered independent component analysis (ST-ICA). The method exploits the central limit theorem to find the directions in the high-dimensional stimulus space of spike-triggered data that are most independent. We demonstrate the improvement of the ST-ICA method over STC analysis using simulated neurons. When tested on data obtained from the H1 neuron in the fly visual system, it predicts a spatial arrangement of functional subunits with adjacent receptive fields. The properties of these subunits strongly resemble the known properties of elementary movement detector inputs to the H1 neuron. Using the ST-ICA method, we derive a model that captures functional and physiological properties of fly motion vision.