Quantitative structure-activity relationship characterization and modeling of length-varying bioactive peptides

Area of Science:

• Computational chemistry and bioinformatics
• Quantitative structure-activity relationship (QSAR) modeling
• Peptidology and drug discovery

Background:

• Traditional quantitative structure-activity relationship (QSAR) methods for bioactive peptides (BAPs) struggle with variable peptide lengths.
• Existing methods like amino acid descriptor (AAD) and one-hot encoding (OHE) result in inconsistent feature vector dimensions.
• Auto-cross covariance (ACC) addresses dimensional inconsistency but suffers from potential feature explosion and low interpretability.

Purpose of the Study:

• To introduce a novel post-peptide characterization strategy, residue descriptor-distance vector (RDDV), to standardize feature vectors for length-varying peptides.
• To develop predictive QSAR models for BAPs using the RDDV strategy on a curated dataset.
• To compare the performance and applicability of RDDV against existing methods like ACC and global descriptor (GD).

Main Methods:

• Developed residue descriptor-distance vector (RDDV) with two subclasses (RDDV(I) and RDDV(II)) for single- and multiple-amino acid descriptor characterizations.
• Applied RDDV to develop QSAR models using the ScBAPqad dataset comprising seven large-scale, length-varying BAP sample sets.
• Conducted comparative analyses of RDDV against ACC and global descriptor (GD) using various characterization strategies and machine learning methods.

Main Results:

• The RDDV strategy effectively scales feature vectors for length-varying peptides, enabling direct QSAR modeling.
• Predictive QSAR models were successfully developed using RDDV on the experimental BAP dataset.
• Comparative analysis provided insights into the applicability domains, limitations, and potential of RDDV relative to ACC and GD.

Conclusions:

• Residue descriptor-distance vector (RDDV) presents a viable and effective alternative to existing methods for peptide characterization in QSAR.
• RDDV facilitates the development of more consistent and potentially more interpretable QSAR models for bioactive peptides.
• The study highlights RDDV's potential to advance computational peptidology and accelerate drug discovery efforts.