Integrating Predictive and Generative Capabilities by Latent Space Design via the DKL-VAE Model
- 1Institute for Materials Science and Center for Nanointegration Duisburg-Essen (CENIDE), University of Duisburg-Essen, Essen, 45141, Germany
- 2Department of Materials Science and Engineering, University of Tennessee, Knoxville, TN, 37996, USA
- 3Pacific Northwest National Laboratory, Richland, WA, 99354, USA
Abstract
We introduce a deep Kernel learning variational autoencoder (DKL-VAE) framework that integrates the generative power of a variational autoencoder (VAE) with the predictive nature of deep Kernel learning (DKL). The VAE learns a latent representation of high-dimensional data, enabling the generation of novel structures, while DKL refines this latent space by structuring it in alignment with target properties through Gaussian process (GP) regression. This approach preserves the generative capabilities of the VAE while enhancing its latent space for GP-based property prediction. We evaluate the framework on two datasets: a structured card dataset with predefined variational factors and the QM9 molecular dataset, where enthalpy serves as the target function for optimization. The model demonstrates high-precision property prediction and enables the generation of novel out-of-training subset structures with desired characteristics. The DKL-VAE framework offers a promising approach for high-throughput material discovery and molecular design, balancing structured latent space organization with generative flexibility.
Related
- Integrating DLVO Theory with Machine Learning for Predictive Toxicology and Multi-Objective Design of Quercetin-Loaded Nanocarriers
- DeepSAP: an integrated generative and predictive deep-learning framework for controllable design of self-assembling peptides.
- Integrating Generative AI and Data Science into General Chemistry Laboratories
- Hybrid Organosilicon Polymers Integrating Latent Si–Si Chromophores and Orthogonal Si–O–C Reactivity
- Integrating Machine Learning for Predictive Modeling of Carbon Sequestration in Mixed Microalgal Systems
BibTeX
@article{slautin2026integrating,
title = {{Integrating Predictive and Generative Capabilities by Latent Space Design via the DKL-VAE Model}},
author = {Boris N. Slautin and Utkarsh Pratiush and Doru Constantin Lupascu and Maxim A Ziatdinov and Sergei V. Kalinin},
journal = {npj Computational Materials},
year = {2026},
month = jul,
doi = {10.1038/s41524-026-02197-5},
url = {https://doi.org/10.1038/s41524-026-02197-5},
note = {CC BY},
}
RIS
TY - JOUR
TI - Integrating Predictive and Generative Capabilities by Latent Space Design via the DKL-VAE Model
AU - Boris N. Slautin
AU - Utkarsh Pratiush
AU - Doru Constantin Lupascu
AU - Maxim A Ziatdinov
AU - Sergei V. Kalinin
PY - 2026
DA - 2026/07/06
JO - npj Computational Materials
DO - 10.1038/s41524-026-02197-5
UR - https://doi.org/10.1038/s41524-026-02197-5
AB - We introduce a deep Kernel learning variational autoencoder (DKL-VAE) framework that integrates the generative power of a variational autoencoder (VAE) with the predictive nature of deep Kernel learning (DKL). The VAE learns a latent representation of high-dimensional data, enabling the generation of novel structures, while DKL refines this latent space by structuring it in alignment with target properties through Gaussian process (GP) regression. This approach preserves the generative capabilities of the VAE while enhancing its latent space for GP-based property prediction. We evaluate the framework on two datasets: a structured card dataset with predefined variational factors and the QM9 molecular dataset, where enthalpy serves as the target function for optimization. The model demonstrates high-precision property prediction and enables the generation of novel out-of-training subset structures with desired characteristics. The DKL-VAE framework offers a promising approach for high-throughput material discovery and molecular design, balancing structured latent space organization with generative flexibility.
SN - 2057-3960
ER -