CPL-Diff: A Diffusion Model for De Novo Design of Functional Peptide Sequences with Fixed Length.

Area of Science:

• Computational biology
• Drug discovery
• Artificial intelligence in medicine

Background:

• Peptides are crucial therapeutic agents for various human diseases.
• Deep learning models show promise for peptide generation but lack length control.
• Sequence length critically influences peptide properties and therapeutic efficacy.

Purpose of the Study:

• Introduce CPL-Diff, a novel diffusion model for generating functional peptide sequences with controlled length.
• Enable precise control over polypeptide sequence length using attention masking.
• Facilitate generation of single-functional peptides based on conditional inputs.

Main Methods:

• Developed CPL-Diff, a diffusion model incorporating attention masking for length control.
• Implemented conditional generation for single-functional peptide design.
• Conducted experiments to evaluate generated peptide quality and properties.

Main Results:

• CPL-Diff generates peptides with lower perplexity and similarity than state-of-the-art models.
• Generated peptides exhibit physicochemical properties comparable to natural sequences.
• Interpretability analysis provides insights into length control mechanisms.

Conclusions:

• CPL-Diff offers precise length control for generating functional therapeutic peptides.
• The model demonstrates superior performance in generating peptides with relevant properties.
• Findings offer theoretical guidance for advanced polypeptide design and drug discovery.