Deterministic parallel global parameter estimation for a model of the budding yeast cell cycle

Abstract

Abstract. Two parallel deterministic direct search algorithms are used to find improved parameters for a system of differential equations designed to simulate the cell cycle of budding yeast. Comparing the model simulation results to experimental data is difficult because most of the experimental data is qualitative rather than quantitative. An algorithm to convert simulation results to mutant phenotypes is presented. Vectors of parameters defining the differential equation model are rated by a discontinuous objective function. Parallel results on a 2200 processor supercomputer are presented for a global optimization algorithm, DIRECT, a local optimization algorithm, MADS, and a hybrid of the two.

Performance Modeling and Analysis of a Massively Parallel Direct—Part 2

“…Additionally, combining DIRECT with other methods has been found to be fairly successful to address this issue (Nelson et al 1998, Zwolak et al 2005, Panning et al 2006. The emphasis of the present work is on the performance analysis of the parallel design, so the original DIRECT with ǫ = 0 and additional stopping rules is considered.…”

“…In addition to the artificial test functions, the cell cycle parameter estimation problems for frog egg extracts (FE) by Zwolak et al (2005) and budding yeast (BY) by Panning et al (2006) are used here as real world applications. The cell cycle modeling aims at decoding the physiological behavior of a living cell.…”

“…Compared to local methods, these global approaches are more likely to discover the global optimum instead of being trapped at local minimum points for complex nonconvex or nonlinear problems with irregular design domain. The DIRECT algorithm (Jones et al 1993) is one such global optimization algorithm that has been applied to large scale engineering design problems such as aircraft design (Baker et al 2000), pipeline design (Carter et al 2001), routing (Bartholomew-Biggs et al 2003), surface optimization (Zhu et al 2002), transmitter placement (He et al 2004b), molecular genetic mapping (Ljungberg et al 2004), and cell cycle modeling (Zwolak et al 2005 andPanning et al 2006). The complexity of these applications ranges from low dimensional with 3-20 variables to high dimensional with up to 143 variables.…”

Design and implementation of a massively parallel version of DIRECT

“…Budding yeast parameter estimator [17] in pDIRECT I work only for a fixed master, so they have to sit idle when the master runs out of work until the next iteration starts.…”

“…Like other global optimization approaches of [4] and [6], DIRECT is being challenged by high-dimensional (≥ 50) problems including nonlinear models for parameter estimation. The present work applies DIRECT to a 143-parameter estimation problem for a budding yeast cell cycle model [17].…”