Cell-penetrating peptides predictors: A comparative analysis of methods and datasets

Area of Science:

• Biochemistry
• Bioinformatics
• Drug Discovery

Background:

• Cell-Penetrating Peptides (CPP) offer therapeutic potential beyond small molecules.
• Numerous predictors exist for identifying and designing novel CPPs.
• Previous performance comparisons of CPP predictors lacked clarity on dataset influence.

Purpose of the Study:

• To systematically evaluate the impact of peptide sequence similarity within datasets on CPP predictor performance.
• To determine whether dataset characteristics, model choice, or descriptors are primary drivers of predictor accuracy.

Main Methods:

• Analysis of CPP predictor performance across datasets with varying degrees of sequence similarity.
• Comparison of classifier outcomes based on the sequence relatedness of positive (CPP) and negative (non-CPP) examples.

Main Results:

• Dataset characteristics, specifically sequence similarity, exert a greater influence on predictor performance than the chosen model or descriptors.
• Classifiers perform well on datasets with low sequence similarity between CPP and non-CPP examples.
• Datasets with high sequence similarity between CPP and non-CPP examples present a significant challenge for accurate prediction.

Conclusions:

• The composition and sequence similarity of training datasets are critical factors in assessing CPP predictor reliability.
• Future evaluations of new CPP predictors should utilize datasets with high sequence similarity to ensure rigorous performance assessment.
• Understanding dataset influence is key to developing more accurate and generalizable CPP identification tools.