Multivariate statistics in analysis of data from the in vitro motility assay

A novel approach is described for classification of filaments as stationary or moving and for extraction of velocity data for smooth actin filament sliding in vitro. Moving and stationary filaments were effectively classified using four discriminating variables in a multivariate statistical analysis. The variables were (1) two different measures of the average filament distance from its starting point, (2) a measure of the variability in sliding direction, and (3) the coefficient of variation (CV) of the frame-to-frame sliding velocity (v(mean)). On the basis of this multivariate analysis we obtained correct classification of 98% of the stationary filaments and 94% of the moving filaments in a cross-validation data set. The same classification functions were useful throughout despite a 10-fold variation in the average sliding velocity in the cross-validation data. Further analysis of motile filaments suggested that the velocity of smooth sliding should, ideally, be obtained from the intercept on the velocity axis of a plot of v(mean) against CV. The velocity, so obtained, was between 10 and 30% (mean 20+/-3%; n=7; p<0.001) higher than if average sliding velocity was obtained for all moving filaments with CV<0.5.