Order of Addition in Mixture‐Amount Experiments

Taha Hasan, Touqeer AhmadPharmaceutical 2025 · DOI

TL;DR

This paper introduces a novel approach to designing experiments for mixtures where not only the proportions of components but also the order in which they are added influences the outcome. They developed a method using the Threshold Accepting (TA) algorithm to create efficient experimental designs that significantly reduce the number of required runs while maintaining high predictive accuracy, making complex mixture optimization more practical.

What Problem Does This Paper Address?

In many scientific and industrial applications, particularly in pharmaceuticals, the final product is a mixture of several components. Traditionally, mixture experiments focus on optimizing the proportions of these components. However, in some cases, the order in which these components are combined can also significantly impact the final product's properties (e.g., drug efficacy, tablet dissolution). This "order-of-addition" (OofA) effect, along with the total amount of the mixture, creates a complex experimental space. Existing methods for OofA mixture experiments can lead to an explosion in the number of required experimental runs as the number of components increases, making them impractical due to cost and time constraints. This paper addresses the challenge of designing efficient experiments that capture both mixture-amount and order-of-addition effects without an overwhelming number of runs.

Methods

The core of this paper lies in developing and applying a methodology for constructing Order-of-Addition (OofA) mixture-amount designs.

  • Modeling: They extend existing mixture-amount models to incorporate OofA effects. This involves defining new variables (PWO factors) that capture the pairwise order of component addition. They propose models that account for interactions between component amounts, total mixture amount, and the order of addition.
  • Design Construction: Full OofA mixture-amount designs can be very large. To overcome this, the authors employ the Threshold Accepting (TA) algorithm. This is a metaheuristic optimization algorithm that iteratively searches for a subset of runs from a larger candidate set. The TA algorithm is particularly useful because it allows for accepting "worse" solutions temporarily, guided by a decreasing threshold, to escape local optima and find a globally better design.
  • Optimality Criterion: The TA algorithm is used to select designs that maximize G-optimality. G-optimality aims to minimize the maximum prediction variance over the entire design space, ensuring reliable predictions across all possible experimental conditions.
  • Efficiency Assessment: The G-efficiency criterion is used to quantify how well a design performs in estimating model parameters and predicting responses, relative to an ideal design.

Datasets

The paper does not introduce new datasets for training or evaluation. Instead, it uses:

  • Simulated datasets: Generated based on established experimental results from the literature (e.g., Cornell [2], Claringbold [24]) to demonstrate the application of their proposed designs and algorithms.
  • Real-world examples: Two case studies are presented:
  • The joint action of three related hormones in groups of mice.
  • A placebo tablet formulation with three components.

These examples leverage existing experimental data and models to showcase the utility of the developed OofA mixture-amount designs.

Key Insights and Ideas

  • Order Matters: The research highlights that for certain applications, the sequence of adding components to a mixture is as crucial as their proportions or total amount for achieving desired outcomes.
  • The Curse of Dimensionality in OofA: The factorial increase in experimental runs with the number of components in full OofA designs is a significant practical barrier.
  • TA Algorithm as a Practical Solution: The Threshold Accepting algorithm proves to be an effective tool for selecting a subset of runs that maintains high G-optimality, making complex OofA mixture-amount experiments feasible.
  • G-Efficiency as a Robust Metric: Focusing on G-optimality ensures that the chosen designs provide stable and reliable predictions across the entire experimental space, which is critical for optimization.
  • Fraction of Design Space (FDS) for Visualization: The FDS plot offers a valuable visual tool to assess the prediction capabilities of a design, showing how much of the design space is covered by low prediction variance.

Key Results

  • The proposed TA algorithm-based designs achieve significantly reduced numbers of experimental runs compared to full OofA designs. For instance, in one example, a 36-run design was selected from a full 63-run design, retaining substantial predictive accuracy.
  • The TA-generated designs demonstrate high G-efficiency, indicating their effectiveness in minimizing prediction variance. For example, a 36-point design achieved a mean G-efficiency of 2451.3, retaining about 51% of the predictive capability of the full design, with a substantially reduced experimental burden.
  • In a simulation study, the TA-reduced design (36 points) achieved a lower median Mean Squared Error (MSE) compared to the full design (63 points), suggesting improved performance in estimating responses.
  • Analysis of real-world examples (hormone study, tablet formulation) showed that both component proportions and the order of addition had significant effects on the response variables.

Limitations and Future Work

  • Computational Cost: While the TA algorithm reduces runs, generating the initial full design and running the algorithm can still be computationally intensive for a very large number of components.
  • Assumptions: The models assume specific forms of interaction between mixture components, total amount, and order of addition. The validity of these models depends on the underlying process.
  • Future Work: The authors suggest extending this work to include:
  • Constrained mixture-amount experiments: Where certain component proportions or amounts are restricted.
  • Orthogonally blocked mixture-amount experiments: To improve efficiency and reduce variability in the presence of blocking factors.
  • Other optimality criteria: Exploring criteria beyond G-optimality.
  • Metaheuristic algorithms that do not require candidate points: To potentially construct even more optimal designs.

Summary updated

Abstract

In a mixture experiment, we study the behavior and properties of m mixture components, where the primary focus is on the proportions of the components that make up the mixture rather than the total amount. Mixture-amount experiments are specialized types of mixture experiments where both the proportions of the components in the mixture and the total amount of the mixture are of interest. In this paper, we consider an Order-of-Addition (OofA) mixture-amount experiment in which the response depends on both the mixture amounts of components and their order of addition. Full mixture OofA designs are constructed to maintain orthogonality between the mixture-amount model terms and the effects of the order of addition. But the number of runs in such full OofA designs increases as m increases. We employ the Threshold Accepting (TA) Algorithm to select an n-row subset from the full OofA mixture design that maximizes G-optimality while minimizing the number of experimental runs. Further, the G-efficiency criterion is used to assess how well the design supports the precise and unbiased estimation of the model parameters. These designs enable the estimation of mixture-component model parameters and the order-of-addition effects. The Fraction of Design Space (FDS) plot is used to provide a visual assessment of the prediction capabilities of a design across the entire design space.
BibTeX
@inproceedings{hasan2025order,
  author = {Taha Hasan and Touqeer Ahmad},
  title = {{Order of Addition in Mixture‐Amount Experiments}},
  booktitle = {Pharmaceutical},
  year = {2025},
  doi = {10.1002/pst.70047},
}
Figure