Target-aware latent diffusion model for design of apoptosis-inducing anticancer peptides.

Area of Science:

• Computational chemistry and drug discovery
• Bioinformatics and structural biology
• Oncology therapeutics

Background:

• Therapeutic peptide design is challenging due to chemical diversity and flexibility.
• B-cell lymphoma-extra large (BCL-xL) is a key anti-apoptotic target in cancer.
• Current generative models lack detailed receptor structural context.

Purpose of the Study:

• To present a target-aware latent diffusion framework (T-LDM) for designing anticancer peptides.
• To condition peptide generation on specific receptor pocket structural information.
• To improve the biological plausibility and target specificity of designed peptides.

Main Methods:

• Developed a target-aware latent diffusion framework (T-LDM).
• Conditioned sequence denoising on receptor pocket context using graph-based structural models.
• Fused conditioning into a U-Net denoiser via modulation and cross-attention.
• Evaluated peptide properties using Jensen-Shannon divergence, perplexity, BLEU, and protein-peptide docking.

Main Results:

• T-LDM generates peptides that are syntactically fluent and physicochemically realistic.
• Explicit pocket conditioning enhances compatibility with the BCL-xL target.
• The model maintains peptide diversity and novelty.
• Ablation studies confirmed the benefit of pocket-aware guidance.

Conclusions:

• Target-aware latent diffusion models offer a promising approach for rational peptide drug design.
• Conditioning generative models on structural context improves therapeutic potential.
• This framework advances the design of targeted peptide therapeutics for oncology.