Decoding the unseen: unsupervised anomaly detection in metal-organic frameworks for discovery beyond the norm

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Discovering novel metal-organic frameworks (MOFs) is crucial for expanding reticular chemistry.
• Ensuring the reliability of MOF datasets is essential for computational materials science.

Purpose of the Study:

• To develop an automated, efficient pipeline for detecting chemically novel or structurally anomalous MOFs.
• To enhance the reliability of MOF databases and facilitate the discovery of new materials.

Main Methods:

• CHEM-AD (Chemically Unusual Metal-organic Frameworks via Autoencoder-based Detection) utilizes 81 engineered descriptors.
• A compact symmetric autoencoder learns typical MOF distributions and identifies anomalies via reconstruction error.
• The pipeline is CPU-efficient, requires no 3D structure fitting, and processes data rapidly.

Main Results:

• CHEM-AD identified 488 anomalous MOFs (1.87%) from 26,025 entries in MOFxDB.
• Anomalies exhibit distinctive topologies, unusual pore metrics (PLD, LCD), and extreme densities.
• Feature attribution highlights connectivity, window geometry, and linker-metal composition as key anomaly drivers.

Conclusions:

• CHEM-AD provides a scalable framework for MOF discovery and database curation.
• The pipeline effectively categorizes anomalies into plausible candidates, chemically resolvable issues, and structural artifacts.
• This approach generalizes to other porous materials, advancing materials informatics.