October 21, 2026

Now available on Nature.com: AI-ready design of realistic 2D materials with Mat3ra-2D

Our paper "AI-ready design of realistic 2D materials and interfaces with Mat3ra-2D" was published online in Scientific Reports (Nature Portfolio) on October 14, 2026. It presents Mat3ra-2D, an open-source framework for designing realistic 2D materials, slabs, interfaces and defects, with 23 notebooks that run in any web browser.

Yuliia Martynenko

| Author

Our new paper, AI-ready design of realistic 2D materials and interfaces with Mat3ra-2D, was published online in Scientific Reports (Nature Portfolio) on October 14, 2026. The authors are Vsevolod Biryukov, Kamal Choudhary and Timur Bazhirov (corresponding author). The paper is Open Access under the CC BY-NC-ND 4.0 license.

Authors

Vsevolod Biryukov – Exabyte Inc. / Mat3ra.com
Kamal Choudhary – Johns Hopkins University: Department of Materials Science and Engineering and Department of Electrical and Computer Engineering
Timur Bazhirov – Exabyte Inc. / Mat3ra.com (corresponding author)

Why realistic structures matter

AI and machine learning models in materials science are mostly trained on ideal bulk crystals. In real devices, surfaces, interfaces and defects dominate, so such models transfer poorly. Realistic structures come from a sequence of choices: orientation, termination, number of layers, vacuum, strain, stacking registry and defect type. They need to be constructed and documented, not just picked from a database.

The Define–Refine–Build workflow for a strain-matched interface: film and substrate slabs are defined, the ZSL analyzer enumerates and ranks commensurate matches, and the interface builder produces the selected structure with strain and shift recorded as metadata. Figure 1 from Biryukov et al., Scientific Reports (2026), CC BY-NC-ND 4.0.

‍

What Mat3ra-2D is

Mat3ra-2D is an open-source framework for the rapid design of realistic 2D materials and related structures: slabs, heterogeneous interfaces, defects and disorder. It rests on two pillars:

Standards for storing and exchanging materials data, with a modular implementation of the core concepts (the M-CODE categorization and ontology).
Transformation workflows written as configuration-builder pipelines that keep provenance and metadata (Define – Refine – Build).

The software stack consists of five packages:

mat3ra-esse: standards and schemas
mat3ra-standata: commonly used example materials
mat3ra-code: abstract classes
mat3ra-made: materials design (slabs, interfaces, strain, defects)
mat3ra-api-examples: Jupyter notebooks

It is interoperable with pymatgen and ASE.

Try it in your browser

The paper comes with 23 reusable notebooks. They cover slabs, monolayers, nanowires, nanoribbons, clusters, heterostructures, interfaces, twisted interfaces, adatoms, islands, terraces, grain boundaries, point defects, passivation, perturbations and cutouts. They run in any web browser via JupyterLite and Pyodide, with no local installation: jupyterlite.mat3ra.com.

The notebooks run in JupyterLite directly in the browser. The left panel lists the notebooks, and the Introduction notebook on the right works as an index with links to each one, organized by M-CODE category. Figure 5 from Biryukov et al., Scientific Reports (2026), CC BY-NC-ND 4.0.

‍

Worked examples in the paper

SrTiO3(110) slabs with different terminations
A Ge(001)/Si(001) interface built without strain matching
A strain-matched Graphene/Ni(001) interface, using the ZSL analyzer to rank candidates by strain and size and recording the strain and shift as metadata

Data

Representative structures are available in the public Materials Explorer on platform.mat3ra.com and are archived on Figshare with full provenance metadata: doi.org/10.6084/m9.figshare.32786562.

Scope and what is next

The paper covers the structure-generation layer that comes before property calculations and ML training. It does not replace validation against experiment. The next steps are validation, scalable dataset design, executable workflows, calculated properties, and dataset generation for downstream AI/ML.

Cite this paper

Biryukov, V., Choudhary, K., Bazhirov, T. AI-ready design of realistic 2D materials and interfaces with Mat3ra-2D. Scientific Reports (Nature Portfolio), published online October 14, 2026. DOI: 10.1038/s41598-026-72825-z.

Read the paper: nature.com/articles/s41598-026-72825-z. This is the early “Article in Press” version; the final edited version will replace it under the same DOI. The work was supported in part by NIST 70NANB24H205.

Your Cart