1997 Volume 33 Issue 8 Pages 812-818
A quantal analysis method is proposed for the study of long-term potentiation (LTP) in hippocampus. This physiological phenomenon has been a leading candidate for mechanisms underlying the memory and learning function. It has been thought that the quantal analysis of miniature excitatory post-synaptic potentials (mEPSPs) would be able to resolve the long standing question of whether LTP is a pre- or post-synaptic event. In practice, however, the analysis is not always useful since the data are usually corrupted by noise. One approach to circumvent this problem would be the application of the maximum entropy noise deconvolution (MEND) method which actively eliminates the noise. MEND is composed of two elements, a maximum likelihood estimator and a maximum entropy. The former is intended to unmask small peaks in the quantal frequency distribution while the latter has a role of smoothing the estimation result, with a Lagrange-multiplier like parameter controlling the balance between the two functions. Our preliminary study has confirmed the feasibility of MEND is fairly good, but the method wastes a large portion of the available data due to its inadequate integration. MEND has two versions, binned and unbinned ones. The unbinned form is more robust, but there is some problem in the way to remove the bins. The unbinned method presented here is a new unbinned form of MEND devised for the quantal analysis of mEPSP recordings. We have improved the unbinned form with the numerical integration to obtain the integral as accurately as possible. The method is aimed at minimizing the noise effect on the estimation performance without wasting the valuable experimental records. Its usefulness is demonstrated by using synthetic as well as real hippocampul mEPSP data. Furthermore, we had applied the improved method to monosynaptic experimental data. Compared to the conventional statistical methods, it gave better results.
