AMPBenchmark

AMPBenchmark provides standardized, bias-aware benchmarking for antimicrobial peptide (AMP) prediction models to assess how negative data sampling strategies affect model performance.

Key Features:

• Comprehensive Model Evaluation: Evaluates 660 predictive models generated from 12 distinct machine learning architectures trained on a single positive dataset and tested using 11 distinct negative data sampling methods.
• Bias Quantification: Identifies and quantifies the influence of negative data sampling on benchmarking outcomes to reveal sampling-induced bias in model performance comparisons.
• Standardized Benchmarking Framework: Provides a standardized evaluation framework that holds positive data constant while varying negative sampling strategies to enable fair comparison of AMP prediction models.

Scientific Applications:

• Novel AMP Discovery: Supports comparative assessment of prediction models to prioritize candidate antimicrobial peptides for experimental validation.
• Drug Development: Informs selection of high-performing predictive models for prioritizing AMP candidates in antibiotic discovery workflows.
• Viral and Cancer Research: Enables evaluation and selection of models that predict AMPs with reported activity against viruses and cancer cells.

Methodology:

Performs systematic analysis of predictive models using a single consistent positive dataset while varying negative data sampling methods, evaluating 660 models produced from 12 machine learning architectures across 11 negative sampling strategies.