MultiCycPermea: accurate and interpretable prediction of cyclic peptide permeability using a multimodal...

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Peptide therapeutics

Background:

• Cyclic peptides offer high binding affinity and low toxicity, making them promising drug candidates for challenging protein targets.
• Poor membrane permeability limits the therapeutic efficacy of cyclic peptides.
• Current methods for assessing peptide permeability are slow and labor-intensive, hindering rapid development.

Purpose of the Study:

• To develop a novel deep learning model for accurate prediction of cyclic peptide membrane permeability.
• To accelerate the assessment of cyclic peptide permeability for drug development.
• To provide insights for designing improved membrane-permeable cyclic peptides.

Main Methods:

• A deep learning model, MultiCycPermea, was developed to predict cyclic peptide permeability.
• The model integrates both 2D (image) and 1D (sequence) structural information of cyclic peptides.
• A substructure-constrained feature alignment module was introduced to harmonize diverse feature types.

Main Results:

• MultiCycPermea significantly improved predictive accuracy, reducing mean squared error (MSE) by 44.83% on the CycPeptMPDB dataset compared to the previous state-of-the-art.
• The model achieved an MSE of 0.16, outperforming Multi_CycGT (0.29).
• Visual analysis tools integrated with MultiCycPermea can elucidate structure-permeability relationships.

Conclusions:

• MultiCycPermea offers an effective and accurate method for predicting cyclic peptide permeability.
• The model provides valuable insights to guide the design of cyclic peptides with enhanced membrane permeability.
• This AI-driven approach facilitates the development of novel, membrane-permeable cyclic peptide therapeutics.