BRS-nonint

BRS-nonint generates balanced negative samples for protein-protein interaction (PPI) prediction by sampling the complement graph to preserve protein degree distributions and mitigate hub-related biases.

Key Features:

• Balanced Negative Sampling: Samples non-interactions from the complement graph to produce a negative dataset whose per-protein degrees mirror those in the positive interaction set.
• Positive set-specific method: Constructs negative datasets by matching each protein's degree in the negative set to its degree in the positive set.
• Bias Mitigation: Reduces biases from hub proteins and degree-related artifacts that can artificially inflate PPI prediction accuracy.
• Feature evaluation capability: Facilitates comparison of feature predictive value, including simple sequence-based features such as amino acid 3mer content and protein domain-based features.

Scientific Applications:

• PPI prediction accuracy assessment: Provides more realistic negative datasets for evaluating the reliability and realism of PPI prediction models.
• Feature validation: Supports assessment of the predictive utility of sequence-based versus protein domain-based features in PPI models.

Methodology:

Uses a positive set-specific sampling procedure that selects non-interactions from the complement graph such that each protein's degree in the negative dataset equals its degree in the positive dataset.