Hybridizing expert descriptors and learned embeddings yields transferable Mat2Vec-* element representations

Abstract

Element representations bridge chemical identity and machine-learning models for materials prediction, but handcrafted descriptors, corpus-derived embeddings, and supervised embeddings are typically treated as separate choices. Here we show that handcrafted and supervised representations encode complementary chemical information and that supervised element geometry can be distilled into reusable element tables (Mat2Vec-*). Across eight structure-based MatBench regression tasks, controlled CGCNN and MEGNet benchmarks span 2,240 models over representation mixtures and latent dimensions. The top-performing models define rotation-invariant cosine-distance relations from which we reconstruct Mat2Vec-*. In composition-only model (CrabNet), these tables replace or augment corpus-derived Mat2Vec, with most configurations improving performance across all four MatBench composition tasks. On the Matbench Discovery WBM benchmark, Mat2Vec-* features further reduce errors for several modern structure-based predictors through lightweight residual correction. Together, these results establish Mat2Vec-* as a compact, transferable supervised chemical prior for materials prediction.

Transformer-inspired architectures make machine-learned electron densities accurate and transferable enough to bypass self-consistency
Jeheon Woo
arXiv29 Oct 2026PreprintarXiv licenceFull textPhysical chemistry Computational physicsarXiv
Selectivity- and Activity-Aware Catalyst Descriptors for CO₂ Hydrogenation on Alloy Nanocatalysts using Machine-Learned Force Fields
Prajwal Pisal, Ondřej Krejčí, Patrick Rinke
arXiv10 Sep 2026PreprintCC BY 4.0Full textPhysical chemistry Computational physicsarXiv
A Transferable Machine-Learned Potential Enabling Large-Scale Copper-Carbon Simulations from Equilibrium to Extreme Conditions
Feiyu Zhou, Wei Qiu, Yan Chen +4
arXiv13 Oct 2026PreprintCC0Full textPhysical chemistry Applied physics & instrumentationarXiv
Electronic configuration vectors provide transferable elemental representations for machine learning across the periodic table
Raghavendra Rajagopalan, Ananth Govind Rajan
ChemRxiv3 Sep 2026PreprintCC BY-NC-NDFull textMaterials & polymersChemRxiv
Learning Electron Density with Virtual Atoms for Transferable Atomistic Representations
Yuzhi Liu, Xinzijian Liu, Chengqian Zhang +3
arXiv10 Oct 2026PreprintarXiv licenceFull textPhysical chemistryarXiv

Cite

BibTeX RIS

BibTeX

@article{liu2026hybridizing,
  title = {{Hybridizing expert descriptors and learned embeddings yields transferable Mat2Vec-* element representations}},
  author = {Xiaotong Liu and Qiong Feng and Haotian Chen and Yuhang Wang and Tao Yang},
  journal = {Communications Chemistry},
  year = {2026},
  month = oct,
  doi = {10.1038/s42004-026-02246-6},
  url = {https://www.nature.com/articles/s42004-026-02246-6},
  note = {CC BY-NC-ND 4.0},
}

RIS

TY  - JOUR
TI  - Hybridizing expert descriptors and learned embeddings yields transferable Mat2Vec-* element representations
AU  - Xiaotong Liu
AU  - Qiong Feng
AU  - Haotian Chen
AU  - Yuhang Wang
AU  - Tao Yang
PY  - 2026
DA  - 2026/10/30
JO  - Communications Chemistry
DO  - 10.1038/s42004-026-02246-6
UR  - https://www.nature.com/articles/s42004-026-02246-6
AB  - Element representations bridge chemical identity and machine-learning models for materials prediction, but handcrafted descriptors, corpus-derived embeddings, and supervised embeddings are typically treated as separate choices. Here we show that handcrafted and supervised representations encode complementary chemical information and that supervised element geometry can be distilled into reusable element tables (Mat2Vec-*). Across eight structure-based MatBench regression tasks, controlled CGCNN and MEGNet benchmarks span 2,240 models over representation mixtures and latent dimensions. The top-performing models define rotation-invariant cosine-distance relations from which we reconstruct Mat2Vec-*. In composition-only model (CrabNet), these tables replace or augment corpus-derived Mat2Vec, with most configurations improving performance across all four MatBench composition tasks. On the Matbench Discovery WBM benchmark, Mat2Vec-* features further reduce errors for several modern structure-based predictors through lightweight residual correction. Together, these results establish Mat2Vec-* as a compact, transferable supervised chemical prior for materials prediction.
SN  - 2399-3669
ER  -

Abstract

Related

Cite