Nonlinear Dimensionality Reduction and Bayesian Optimization Tested for Biobased Foam Formulation Optimization

A scientific article titled 'Nonlinear dimensionality reduction and Bayesian optimization for accelerating design of materials' was published on 09 May 2026. com reported. The study evaluates nonlinear dimensionality reduction methods t-SNE and UMAP in comparison with PCA for biobased foam optimization.

It used an existing dataset comprising 26 distinct methylcellulose-cellulose fiber foam formulations with rheological and mechanical measurements. Optimizing biobased foam formulations is challenging because experiments are costly and fast-to-measure surrogate properties occupy high-dimensional spaces.

Bayesian optimization with Gaussian process regression is used to guide data-efficient searches for biobased foam formulations.

A Gaussian process was trained on low-dimensional representations of rheological observables from the 26-formulation dataset. Bayesian optimization was applied in a single-step, non-sequential manner on the fixed dataset to identify high-yield-stress regions. A second Gaussian process maps foam-formulation compositions to the reduced rheological coordinates.

Across all dimensionality reduction methods, Bayesian optimization identifies similar high-performing formulations achieving yield stress values comparable to the experimentally validated optimum. PCA is used as a baseline due to its deterministic and hyperparameter-free nature. Nonlinear methods (t-SNE and UMAP) can achieve comparable performance to PCA when appropriately tuned.

The study demonstrates that nonlinear DR-assisted BO provides a data-efficient framework for optimizing rheology-governed soft-matter materials. The keywords are: Dimensionality reduction (DR), Principal component analysis (PCA), T-distributed stochastic neighbor embedding (t-SNE), Uniform manifold approximation and projection (UMAP), Gaussian process (GP), Bayesian optimization (BO).

The authors are Muhammad Osman Nadeem Farooqui, Isaac Y. Miranda-Valdez, Tero Mäkinen, Juha Koivisto, and Mikko J. Alava. O. Box 15600, 00076, Espoo, Finland. Muhammad Osman Nadeem Farooqui is the corresponding author.

The authors acknowledge computational resources provided by the Aalto University School of Science “Science-IT” project. I. M. V. thanks the Vilho, Yrjö, and Kalle Väisälä Foundation of the Finnish Academy of Science and Letters for personal funding.

T. M. acknowledges funding from the Academy of Finland (grants 341440 and 346603). The authors acknowledge funding from the FinnCERES flagship (151830423). The authors acknowledge funding from Business Finland (grant numbers 210129, 211835, 211909, 211989).

The authors acknowledge funding from the Future Makers program and the Finnish Cultural Foundation. The authors acknowledge funding from the EU ARCHIBIOFOAM project under the European Union’s Horizon Europe research and innovation programme, grant agreement No 101161052. The authors declare no competing interests.

M. O. N. F. is deeply grateful to Dr. Sadia Zafar for her thoughtful suggestions that helped improve the clarity of the manuscript. 0 International License.