CycPeptPPB

CycPeptPPB predicts the plasma protein binding rate (%PPB) of cyclic peptides to support pharmacokinetic evaluation in drug discovery.

Key Features:

• Deep learning-based modeling: Uses deep learning to predict %PPB for cyclic peptides.
• Residue-level decomposition: Decomposes the macrocycle ring into individual residues for residue-level feature analysis based on sequence information.
• Circular data augmentation: Applies circular data augmentation to represent the inherent circularity of cyclic peptides.
• CyclicConv circular convolution: Implements a novel circular convolution method called CyclicConv to account for peptide circularity in feature extraction.
• Local structural resolution: Captures local structural nuances that influence PPB through residue-level features.
• Performance: Achieves MAE of 4.79% and correlation coefficient R of 0.92 on public drug datasets, outperforming traditional methods (MAE 15.08%, R 0.63).

Scientific Applications:

• Pharmacokinetic evaluation: Inform ADME and pharmacokinetic assessment of cyclic peptide candidates via %PPB prediction.
• Cyclic peptide drug development: Guide design and optimization of cyclic peptide therapeutics by assessing PPB liabilities.
• Targeting challenging proteins: Support development of cyclic peptides intended to target proteins inaccessible to conventional small molecules or antibodies by characterizing PPB.

Methodology:

Employs deep learning on sequence-based residue-level features by decomposing macrocycles into residues, applying circular data augmentation, and using a circular convolution operation (CyclicConv).