PepLM-GNN Ablation T-Test P-Values for Peptide Interaction

17244 peptide samples underpin the statistical t-test results comparing a complete PepLM-GNN model against four ablation variants (A, B, C, D). The table reports p-values for accuracy (ACC) from five-fold cross-validation, indicating the significance of each removed module. Authored by Ke Yan, this data supports analysis of model component contributions in peptide-protein interaction prediction.

Use Cases

• Compare the statistical significance (p-value) of ablation variant A versus the complete model to assess the contribution of the pretrained and classification modules.
• Analyze p-values across ablation variants B, C, and D to determine which graph neural network type (GAT, GCN, GIN) most significantly impacts model accuracy.
• Use the p-values and the 0.05 significance threshold to validate which ablation variants cause a statistically significant drop in the ACC metric.
• Benchmark the reported p-values against other ablation studies in hybrid graph network research for peptide interaction prediction.

Strengths

• Statistical tests are based on a substantial benchmark dataset of 17244 peptide samples (8622 positive and 8622 negative pairs).
• Results are derived from a rigorous five-fold cross-validation procedure, improving reliability.
• Data clearly defines a 0.05 p-value threshold for determining statistical significance of model components.

Limitations

• The dataset is very small (5.5 KB), containing only summary statistical results without the underlying raw performance data.
• Unknown row and column counts limit understanding of the table's structure and the granularity of the reported p-values.
• The temporal and geographic context of the underlying 17244 peptide samples is not provided.

Provenance

Source

Ke Yan, via figshare.

Collection Method

Derived from ablation experiments and statistical t-tests on a five-fold cross-validation set of the PepLM-GNN model.

Freshness

Last updated March 24, 2026.