Scholarlink

<p>Stochastic simulations underpin computational research in fields from systems biology and epidemiology to finance and physics-informed machine learning, yet their reproducibility is hard to quantify because each run yields different outcomes. Existing practices to improve computational reproducibility focus on sharing code, simulation seeds, data, and software environments but do not ensure independent reproduction of results, limiting scientific rigor. Here we introduce the Empirical Characteristic Function Equality Convergence Test (EFECT), a universal computational framework for quantifying the statistical reproducibility of stochastic simulations based on empirical characteristic functions. EFECT defines a normalized EFECT error that measures distributional differences between two sets of simulation outputs over a model- and scale-independent range, and an EFECT convergence point that specifies the minimum sample size required to achieve a desired reproducibility threshold at a chosen significance level. EFECT applies to any simulation whose results can be represented as bounded, real-valued data, regardless of modeling formalism or source of stochasticity. To facilitate widespread adoption and data exchange, we implemented EFECT in an open-source software library and a compact, machine-readable standardized report. We rigorously evaluated the framework across more than 40 test cases spanning stochastic differential equations, agent-based models, Boolean networks, and partial differential equations. Applying EFECT to published models in pandemic epidemiology, financial stochastic volatility, and physics-informed neural networks reveals that commonly used sample sizes are often underpowered: substantial parameter differences can go undetected unless thousands of replicates are simulated. By providing a domain-agnostic statistical reproducibility metric, reusable software libraries, and a standardized report format, EFECT offers a practical framework for embedding quantitative reproducibility and replicability checks into stochastic simulation workflows across data-intensive disciplines. Thus, EFECT delivers the capability to make scientific inferences and policy decisions on the basis of reliable, reproducible stochastic simulations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

EFECT enables reliable and reproducible stochastic simulations across disciplines

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • T. J. Sego,
  • Matthias König,
  • Luis L. Fonseca,
  • Dilan Pathirana,
  • Frank T. Bergmann,
  • Hernán E. Grecco,
  • Mauro Silberberg,
  • Subhasis Ray,
  • Baylor Fain,
  • Adam C. Knapp,
  • Krishna Tiwari,
  • Henning Hermjakob,
  • Herbert M. Sauro,
  • James A. Glazier,
  • Reinhard C. Laubenbacher,
  • Rahuman S. Malik-Sheriff

摘要

Stochastic simulations underpin computational research in fields from systems biology and epidemiology to finance and physics-informed machine learning, yet their reproducibility is hard to quantify because each run yields different outcomes. Existing practices to improve computational reproducibility focus on sharing code, simulation seeds, data, and software environments but do not ensure independent reproduction of results, limiting scientific rigor. Here we introduce the Empirical Characteristic Function Equality Convergence Test (EFECT), a universal computational framework for quantifying the statistical reproducibility of stochastic simulations based on empirical characteristic functions. EFECT defines a normalized EFECT error that measures distributional differences between two sets of simulation outputs over a model- and scale-independent range, and an EFECT convergence point that specifies the minimum sample size required to achieve a desired reproducibility threshold at a chosen significance level. EFECT applies to any simulation whose results can be represented as bounded, real-valued data, regardless of modeling formalism or source of stochasticity. To facilitate widespread adoption and data exchange, we implemented EFECT in an open-source software library and a compact, machine-readable standardized report. We rigorously evaluated the framework across more than 40 test cases spanning stochastic differential equations, agent-based models, Boolean networks, and partial differential equations. Applying EFECT to published models in pandemic epidemiology, financial stochastic volatility, and physics-informed neural networks reveals that commonly used sample sizes are often underpowered: substantial parameter differences can go undetected unless thousands of replicates are simulated. By providing a domain-agnostic statistical reproducibility metric, reusable software libraries, and a standardized report format, EFECT offers a practical framework for embedding quantitative reproducibility and replicability checks into stochastic simulation workflows across data-intensive disciplines. Thus, EFECT delivers the capability to make scientific inferences and policy decisions on the basis of reliable, reproducible stochastic simulations.