Abstract
We assessed the performance of three spike train metrics, Dspike (Victor and Purpura, 1996), the vector product metric of Schreiber et al (2003), and a rate code metric (Dcount), on computer simulated spike trains and single unit recordings from area MT of awake macaques. During the experiment, the animal responded to a direction change on a moving random dot pattern (RDP) located inside the cells' receptive field. In different trials we changed the RDP direction and contrast. We collected data from 102 single neurons in area MT of two animals and compared the similarity of spike trains elicited by different stimulus attributes using the three metrics. We then calculated stimulus clustering information entropy of the spike trains subject to the different metrics. We also derived the performance an ideal observer using the spike trains metrics to perform a 2AFC task, and applied the model to the simulated and physiological data. The computer simulation results demonstrated that overall, the vector product metric was most sensitive to spike timing changes in frequency, but that Dspike was most sensitive to spike timing codes in which the power spectra of the spike trains are the same. There was significantly higher clustering entropy in the recorded spike trains for both direction (mean H=.16 vs H=.35) and contrast (mean H=0.25 vs H=0.46) using Dspike, than a rate code. This was reflected in the better performance of the ideal observer using the Dspike metric instead of a rate code for both directions (mean Pc=0.63 vs 0.71) and contrast (mean Pc=0.65 vs 0.74). The vector product metric yielded intermediate values. Combined these results suggest a potential functional advantage to spike time coding by MT neurons which is not based on temporal frequency changes.