Multi_CycGT: A Deep Learning-Based Multimodal Model for Predicting the Membrane Permeability of Cyclic Peptides

Area of Science:

• Computational chemistry and chemical biology
• Drug discovery and medicinal chemistry

Background:

• Cyclic peptides show promise as therapeutics due to high binding affinity, low toxicity, and ability to target challenging proteins.
• Limited membrane permeability hinders cyclic peptides from reaching intracellular targets, necessitating costly and time-consuming experimental validation.

Purpose of the Study:

• To develop an accurate and efficient computational model for predicting cyclic peptide membrane permeability.
• To accelerate the drug design process for cyclic peptides targeting intracellular proteins.

Main Methods:

• Proposed Multi_CycGT, a multimodal deep learning model integrating graph convolutional networks (GCN) and transformers.
• The model extracts one- and two-dimensional features to predict cyclic peptide membrane permeability.
• Extensive benchmarking and validation on external datasets were performed.

Main Results:

• Multi_CycGT achieved state-of-the-art performance with an average accuracy of 0.8206 and an AUC of 0.8650.
• The model demonstrated strong generalization capabilities on unseen data.
• This represents the first deep learning application for predicting cyclic peptide membrane permeability.

Conclusions:

• The Multi_CycGT model offers a significant advancement in predicting cyclic peptide membrane permeability.
• This computational tool can expedite the development of novel cyclic peptide-based therapeutics.
• Facilitates medicinal chemistry and chemical biology research by reducing experimental bottlenecks.