Innovation Density: A Metric for Evaluating Generative Peptide Models

The first file is an article. The second file is a Python code that allows you to perform calculations using the method described in the article. Modern generative models (LigandForge, RFDiffusion, BindCraft, and others) enable peptide generation at record speed, but their comparative analysis is hampered by the lack of a unified metric that evaluates the ability to generate new structural motifs, rather than reproducing the training set. This paper proposes the "innovation density" metric (ρ_innov), which measures the number of new structural clusters per million generated peptides. The proposed metric is model-independent and can be applied to any generative approach. A formal definition is provided, δ thresholds, and clustering methods are discussed. This metric has the potential to become a standard for evaluating generative peptide models in Target 2035 and similar projects.