next up previous
Next: MemExp: Fits by Distributed

MemExp Documentation

Peter J. Steinbach
Center for Molecular Modeling
Center for Information Technology
National Institutes of Health

Recovering Distributed Lifetimes and Discrete Exponentials from Kinetics
Documentation for MemExp version 5.0

The program MemExp uses the maximum entropy method (MEM) and either nonlinear least squares (NLS) or maximum likelihood (ML) fitting to analyze a general time-dependent signal in terms of distributed and discrete lifetimes. One or two distributions of effective log-lifetimes, $g(log \tau)$ and $h(log \tau)$, plus an optional polynomial baseline (up to a cubic) can be extracted from the data. The h distribution is used to account for signals opposite in sign to those described by the g distribution when analyzing data that rise and fall. Both distributions are obtained numerically from the data and are not restricted to any functional form. MemExp also performs a series of fits by discrete exponentials in which exponentials are added one at a time, based on the recommended MEM distribution. The amplitude and log-lifetime of each exponential, plus any optional baseline parameters utilized, are varied using either NLS (for Gaussian noise) or ML (for Poisson noise) fitting. MemExp automatically recommends one distributed and one discrete description of the kinetics as optimal. The graphical summary plotted by MemExp permits a thorough evalutaion of the results. Multiple MEM 'prior models' are supported, facilitating a comprehensive analysis of the kinetics data. MemExp was written in FORTRAN77 and has been compiled for use on Linux, Windows, and Mac machines. Graphical output is in PostScript format.

New to version 5.0: Zero-time shift can now be accounted for when deconvolving an instrument response function (IRF). Improved automation of fits by discrete exponentials: Each fit is now initialized based on the recommended distribution, not on different iterations during convergence of MEM. The recommended discrete fit is now chosen based, in part, on the Akaike information criterion (AIC). A fixed, constant baseline can now be set on the command line in 'simple' mode. Improved initial estimation of time-dependent standard errors now obtained from the differences between the raw data and the data after smoothing. Background now automatically subtracted from IRF.

Recommendation: First, try MemExp in simple mode, as demonstrated in the test cases below. To change the parameter values assumed in simple mode, the file named data_file.mem output in simple mode can be edited for a second analysis using auto mode. Use of allin1 mode may be necessary if the standard errors are to be estimated, and the two estimates obtained in auto mode haven't converged satisfactorily.

Recommendation: In many cases, it may be advantageous to set IBIGF = 3 and LTCUT to a value of log $\tau$ that excludes the short-lifetime peaks that may appear with a uniform prior (as in example 4 below). However, for kinetics with both sharp and broad phases, consider IBIGF = 1 and 'differential blurring' (as in example 2 below).

References: Please refer to the following papers when publishing results obtained using MemExp.

P.J. Steinbach, R. Ionescu, and C.R. Matthews. Analysis of Kinetics using a Hybrid Maximum-Entropy / Nonlinear-Least-Squares Method: Application to Protein Folding. (2002) Biophys. J. 82: 2244-2255.

P.J. Steinbach. Filtering Artifacts from Lifetime Distributions when Maximizing Entropy using a Bootstrapped Model. (2012) Analytical Biochemistry 427: 102-105.

Return to MemExp home page.

next up previous
Next: MemExp: Fits by Distributed
Peter J. Steinbach 2017-10-30