PRIMED Model Performance Benchmarks for DNA-Binding Residue Prediction

PRIMED is a machine learning framework for predicting DNA-binding residues using protein language models. The model was evaluated on benchmark datasets Test-46, Test-129, and Test-10K, achieving AUC scores up to 0.93.

Use Cases

• Benchmark new DNA-binding residue predictors against PRIMED's reported AUC and MCC scores on Test-46 and Test-129 datasets.
• Analyze model generalizability using performance metrics from the Test-10K dataset curated from UniProtKB/Swiss-Prot.
• Study the integration of protein representations from ESM-2, ESM-3, and ESM-C language models for residue-level prediction tasks.

Strengths

• Model performance is quantified with specific metrics, including an AUC of 0.92 and MCC of 0.64 on the Test-46 benchmark.
• Evaluation spans three distinct benchmark datasets (Test-46, Test-129, Test-10K) to assess transferability.
• Framework integrates representations from three established protein language models: ESM-2, ESM-3, and ESM-C.

Limitations

• The dataset is a 390.7 KB document describing the model and results, not a structured data table for direct analysis.
• Underlying protein sequence or residue-level data for the benchmarks is not included in this file.

Provenance

Source

figshare, author Luoshu Zhang.

Collection Method

Model development and benchmarking using datasets from prior studies and a new Test-10K set curated from UniProtKB/Swiss-Prot.

Time Range

null

Freshness

Last updated March 2026.

Geography

null