Statistical Design Methods for Experimental Optimization

Statistical design methods provide systematic approaches for planning experiments to explore complex systems efficiently. These methods encompass strategies such as factorial and fractional factorial designs for screening a modest number of factors, response surface methodologies (RSM) for modelling and optimising continuous variables, and optimal designs that seek to maximise information under cost constraints. In recent years, advances in algorithmic design search, composite strategies combining screening and optimisation phases, and the integration of Bayesian and machine-learning techniques have expanded the toolkit for experimentalists across disciplines. These approaches have been applied to fields as varied as chemical formulation, materials science, biotechnology and manufacturing, delivering substantial reductions in experimental runs while preserving precision in estimating main effects, interactions and nonlinear responses. The rising complexity of modern research questions—driven by high-throughput platforms and big data—has catalysed the development of supersaturated and definitive screening designs, which enable the rapid identification of critical factors from large pools with minimal resource expenditure. Concurrently, the advent of efficient search algorithms and novel composite frameworks has improved the robustness and efficiency of designs, ensuring global significance in translational and fundamental research.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Innovations in composite design for response surface methodology have introduced new hybrid frameworks that integrate orthogonal matrices with axial components to enhance the efficiency of definitive screening designs. These composite designs deliver superior D-values and robustness compared with traditional central composite and Box–Behnken arrangements, yielding more precise curvature estimates in multidimensional optimisation. In the realm of supersaturated designs, Bayesian D-optimal augmentation techniques have been proposed to incorporate newly identified factors into existing small-run experiments. By optimising the determinant of the posterior information matrix, these augmented designs maintain orthogonality and identifiability, allowing for cost-effective follow-up studies that refine factor screening while controlling for latent effects. Furthermore, progress in algorithmic search methods has been demonstrated through bit-parallel tabu search algorithms engineered to find E(s²)-optimal and minimax-optimal supersaturated designs. By mapping design search to resolvable block structures, these algorithms locate designs that achieve sharp lower bounds on average column correlations, thereby minimising confounding in unreplicated experiments and facilitating rapid variable selection in high-dimensional settings.

Statistical Design Methods for Experimental Optimization publication trend

The graph below shows the total number of articles in statistical design methods for experimental optimization across all publications each year (not limited to Nature Index journals).

Technical terms

Factorial design: An experimental strategy in which all possible combinations of factor levels are tested to assess main effects and interactions.

Fractional factorial design: A subset of a full factorial design selected to reduce runs while preserving estimation of low-order effects.

Response surface methodology (RSM): A collection of statistical and mathematical techniques for modelling and optimisation of responses influenced by multiple continuous factors.

Definitive screening design (DSD): A three-level screening design that allows independent estimation of main effects and selected interactions with minimal runs.

Supersaturated design (SSD): An experimental design in which the number of factors exceeds the number of runs, used for rapid initial factor screening.

D-optimality: A criterion for experimental design that seeks to maximise the determinant of the information matrix, enhancing precision of parameter estimates.

Bayesian D-optimality: An extension of D-optimality incorporating prior information into the design criterion to improve estimation under uncertainty.

References

1. New approaches on composite designs for Response Surface Methodology. PLOS ONE (2024).
2. A Method for Augmenting Supersaturated Designs with Newly Added Factors. Mathematics (2022).
3. A Bit‐Parallel Tabu Search Algorithm for Finding E(s2)‐Optimal and Minimax‐Optimal Supersaturated Designs. Computational and Mathematical Methods (2023).