[논문 리뷰] Learning to Dock: Geometric Deep Learning for Predicting Supramolecular Host-Guest Complexes

Zidi Wang, Tao Zhang외 8명|arXiv (Cornell University)|2026. 01. 18.

Supramolecular Chemistry and Complexes인용 수 0

한 줄 요약

논문은 DeepHostGuest를 소개합니다. 이것은 기하학적 딥러닝 프레임워크로서 호스트-게스트 결합 구성을 예측하고 다양한 범주의 초분자계 시스템에서 정확한 결합 자유에너지 계산을 가능하게 합니다.

ABSTRACT

Predicting non-covalent host-guest recognition remains challenging due to the complex interplay of electrostatics, dispersion, and steric effects, and the limited transferability of existing docking approaches to synthetic supramolecular systems. Here we present DeepHostGuest, a geometric deep-learning framework that learns generalizable recognition principles directly from experimentally resolved host-guest structures. Hosts are encoded as electrostatic surfaces and guests as molecular graphs, enabling transferable learning across diverse supramolecular systems. DeepHostGuest achieves high-accuracy predictions (RMSD $\leq 2$ Angstrom for 80.8% of test cases), substantially outperforming classical docking without case-specific tuning. Notably, the method generalizes beyond its training domain to crystalline sponge systems, accurately capturing the binding of large amphiphilic molecules within metal-organic cages. Beyond predicting binding conformations, the structures generated by DeepHostGuest serve as a reliable basis for accurate binding free-energy calculations. Density Functional Theory (DFT)-calculated affinities correlate well with experiment, enabling structure-property relationships across 876 host-guest complexes spanning 34 host families. Interpretable feature analysis reveals that binding strength arises from a cooperative interplay of host polarity, guest hydrophobicity, and geometric complementarity, with distinct design regimes across supramolecular classes. Together, these results establish data-driven molecular recognition as a practical route to predictive supramolecular design, enabling high-throughput virtual screening and rational optimization of functional host-guest systems.

연구 동기 및 목표

• Predictive supramolecular design for non-covalent host–guest recognition을 위한 동기 부여.
• Develop a transferable learning framework that generalizes across diverse supramolecular systems.
• Encode hosts as electrostatic surfaces and guests as molecular graphs to capture interactions.

제안 방법

• Encode hosts as electrostatic surfaces to capture polarity and charge distributions.
• Encode guests as molecular graphs to represent topology and connectivity.
• Use geometric deep learning to learn recognition principles from experimentally resolved structures.
• Train to predict binding conformations and generate structures suitable for binding-energy calculations.
• Assess transferability to crystalline sponge systems and large amphiphilic guests within metal-organic cages.

실험 결과

연구 질문

• RQ1 Can a geometric deep-learning model learn general principles of host–guest recognition from experimental structures?
• RQ2 To what extent does encoding hosts as electrostatic surfaces and guests as graphs enable transferability across diverse supramolecular classes?
• RQ3 How well do generated docked structures support accurate binding free-energy calculations?
• RQ4 Is the method capable of handling crystalline sponge systems and large amphiphilic guests in MOF cages?
• RQ5 What are the interpretable factors contributing to binding strength across different host families?

주요 결과

• DeepHostGuest achieves RMSD ≤ 2 Angstrom for 80.8% of test cases.
• Outperforms classical docking without case-specific tuning.
• The method generalizes to crystalline sponge systems and large amphiphilic guests in MOF cages.
• DFT-calculated affinities correlate well with experiment across 876 host–guest complexes spanning 34 host families.
• Interpretable analysis shows binding strength from cooperative host polarity, guest hydrophobicity, and geometric complementarity.