Skip to content

model_1.0.0_benchmark.md

Latest commit

 

History

History
30 lines (22 loc) · 2.31 KB

model_1.0.0_benchmark.md

File metadata and controls

30 lines (22 loc) · 2.31 KB

Protenix v1.0.0 Series Benchmarks

This document provides detailed benchmark performance results and access to the evaluation datasets used for Protenix v1.0.0.

📂 Benchmark Data & Results

The complete benchmark results are available for download. Each package contains detailed inference metrics for individual predictions as well as aggregated performance reports.

Dataset Size Description Download
PXM-2024 ~32 MB Benchmark results for structures deposited in 2024. Link
PXM-2025 ~28 MB Benchmark results for structures deposited in 2025. Link
PXM-2025-H2 ~18 MB Benchmark results for structures deposited in 2025 (second half). Link
PXM-22to25-Ab-Ag ~18 MB Focused benchmark on Antibody-Antigen complexes (2022-2025). Link
PXM-22to25-Ligand ~6 MB Focused benchmark on Protein-Ligand complexes (2022-2025). Link
PXM-Legacy ~15 MB Historical benchmark sets used in previous development cycles. Link
foldbench ~11 MB Evaluation results on the external FoldBench dataset. Link
runs-n-poses ~15 MB Benchmark focused on protein-ligand and multi-pose evaluation. Link

How to Analyze Results

To use these benchmark results:

  1. Download: Obtain the relevant .tar.gz file using the links provided above.
  2. Extract: Decompress the file using tar -xzvf <dataset_name>.tar.gz.
  3. Data Structure:
    • Inference Metrics: Detailed per-prediction metrics (e.g., LDDT, RMSD, confidence scores) provided in CSV format for each sample.
    • Aggregated Results: Summary reports providing statistical overviews across the entire dataset for quick performance assessment.