APCyc: Property-Informed Design of Cyclic Peptides via Automated Cyclization

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, extended

Summary

APCyc is a novel target-aware latent diffusion framework designed for de novo cyclic peptide generation, explicitly addressing limitations of existing models in handling cyclization-specific constraints and multi-property optimization. Accepted at KDD 2026, APCyc automates the inference of pocket-adaptive cyclization linkage types and sites directly from the receptor context. It incorporates an expanded residue vocabulary and leverages Bayesian posterior guidance to jointly optimize multiple drug-relevant properties, including affinity, permeability, protease resistance, solubility, and immunogenicity. Experimental results show APCyc achieves the best permeability proxy (0.107) and protease-resistance score (-1.474) among tested methods. It also demonstrates leading structural quality with the best Rosetta total score (-758.545) and consistency (0.971), while maintaining strong binding affinity. The framework's source code is available on GitHub.

Key takeaway

For Machine Learning Engineers developing peptide therapeutics, APCyc offers a robust framework to overcome challenges in cyclic peptide design. You should consider integrating this approach to automate cyclization topology selection and simultaneously optimize multiple drug-relevant properties like permeability and protease resistance. This can accelerate the discovery of candidates with desired therapeutic profiles, but always complement computational predictions with expert assessment and experimental validation due to inherent model uncertainties.

Key insights

APCyc automates cyclic peptide design by inferring cyclization topology and optimizing multiple drug-like properties via Bayesian guidance.

Principles

• Cyclization topology should be a target-conditioned design variable, not a fixed heuristic.
• Explicitly modeling cyclization sites and linkage types improves generative models.
• Bayesian posterior guidance enables controllable multi-objective property optimization in latent diffusion.

Method

APCyc uses a latent diffusion model with an expanded residue vocabulary and cyclization-aware embeddings. It injects topology signals into an AM-EGNN denoiser and applies Bayesian posterior guidance from energy-based surrogates for multi-property optimization.

In practice

• Extend residue vocabulary to distinguish cyclized from non-cyclized residues.
• Use gradient-based guidance in diffusion models for multi-objective optimization.
• Decompose cyclization topology into residue-level and pairwise linkage signals.

Topics

• Cyclic Peptides
• De Novo Design
• Latent Diffusion Models
• Drug Discovery
• Multi-Property Optimization
• Bayesian Posterior Guidance

Code references

• HKUSTGZ-ML4Health-Lab/APCyc

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Sample-efficient generative molecular design using memory manipulation — A novel framework named Saturn, which applies the Mamba architecture to generative molecular design, significantly improves sample efficiency for drug discovery.
• Why We're Not Close to AI Drugs Yet | Isomorphic Labs — Drug design comprises multiple distinct ML problems, not a single one, despite AlphaFold 2's protein structure prediction success.
• Graph-Theoretic Models for the Prediction of Molecular Measurements — Enhanced classical graph-theoretic models can rival deep learning for molecular property prediction with significantly lower computational cost.
• Contrastive Geometric Learning Unlocks Unified Structure- and Ligand-Based Drug Design — ConGLUDe unifies structure- and ligand-based drug design via contrastive geometric learning, enabling diverse drug discovery tasks.
• CP-Agent: Context-Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations — CP-Agent uses an MLLM and context-aware image-metadata embedding to generate interpretable rationales for drug-induced cell changes.
• Bridging the phenotype-target gap for molecular generation via multi-objective reinforcement learning — ExMolRL combines phenotype-guided generation with target-aware reinforcement learning for de novo drug discovery.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.