Quasi-Monte Carlo methods applied to tau-leaping in stochastic biological systems

TL;DR

This paper explores the integration of quasi-Monte Carlo methods with tau-leaping for simulating stochastic biological systems, demonstrating improved efficiency over traditional Monte Carlo approaches despite unique convergence challenges.

Contribution

It introduces a novel combination of quasi-Monte Carlo techniques with tau-leaping for chemical reaction network simulations, highlighting their potential for enhanced accuracy.

Findings

Quasi-Monte Carlo methods often outperform Monte Carlo in variance reduction.

The convergence behavior is complex due to the discrete nature of chemical models.

Performance depends on problem-specific factors, as shown in quadrature tests.

Abstract

Quasi-Monte Carlo methods have proven to be effective extensions of traditional Monte Carlo methods in, amongst others, problems of quadrature and the sample path simulation of stochastic differential equations. By replacing the random number input stream in a simulation procedure by a low-discrepancy number input stream, variance reductions of several orders have been observed in financial applications. Analysis of stochastic effects in well-mixed chemical reaction networks often relies on sample path simulation using Monte Carlo methods, even though these methods suffer from typical slow $\mathcal{O}(N^{-1/2})$ convergence rates as a function of the number of sample paths $N$. This paper investigates the combination of (randomised) quasi-Monte Carlo methods with an efficient sample path simulation procedure, namely $\tau$-leaping. We show that this combination is often more effective than traditional Monte Carlo simulation in terms of the decay of statistical errors. The observed convergence rate behaviour is, however, non-trivial due to the discrete nature of the models of chemical reactions. We explain how this affects the performance of quasi-Monte Carlo methods by looking at a test problem in standard quadrature.