Pathos Plm Embeddings

Precomputed protein language model embeddings for missense substitutions and wild-type residues in 20,416 human SwissProt proteins, used by the PATHOS tool. The dataset was created by DSIMB and was last updated on May 15, 2026. It is designed to predict the pathogenicity of missense mutations.

Use Cases

• Predicting pathogenicity of missense mutations based on precomputed protein language model embeddings.
• Benchmarking variant effect prediction algorithms against the embeddings for 20,416 human proteins.
• Training machine learning models for protein function prediction using the provided mutation and wild-type residue embeddings.

Strengths

• Covers 20,416 human SwissProt proteins, providing a broad scope for analysis.
• Includes embeddings for both generated missense substitutions and wild-type residues, enabling comparative studies.
• Associated with a published research paper (DOI: 10.1016/j.ailsci.2026.100165), suggesting academic validation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

DSIMB

Collection Method

Precomputed from protein language models, likely for use with the PATHOS tool.

Freshness

Last updated 2026-05-15 17:27:19; freshness should be verified.