Machine Learning

Vilya-1: An all-atom foundation model for macrocycle structure prediction and design

Pascal Sturmfels, Milad Salem, Naozumi Hiranuma, Stephen Rettie, Xiaoliang Pan

Featured July 16, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new AI model, Vilya-1, can accurately predict the 3D shapes of complex ring-shaped molecules and guess their drug-like properties, helping scientists design better medicines faster.

In depth

The paper introduces Vilya-1, a deep learning model that significantly advances macrocycle structure prediction and design. It uses a novel all-atom representation and a unified transformer architecture, enabling it to generalize across diverse chemical spaces, including non-canonical residues and small molecules. This approach allows for accurate sampling of biologically relevant conformations and serves as a foundation for downstream tasks like property prediction and generative design.

Key Takeaways

1
Vilya-1 achieves state-of-the-art accuracy in sampling experimentally observed macrocycle structures, outperforming existing physics-based and deep learning methods, especially for chemically diverse systems.
2
The model's all-atom representation and unified architecture enable robust generalization across a broad range of macrocycle chemistries and sizes, including non-canonical residues and small molecules.
3
Vilya-1 functions as a foundation model, supporting downstream applications such as efficient confidence estimation for conformer ranking and high-resolution prediction of drug-like properties like permeability.

Conceptual Flow

HIGH LEVEL

Methodology: How Vilya-1 Works

Vilya-1 takes detailed molecule information, processes it with a smart AI brain, and then creates many possible 3D shapes for the molecule.

Atom Details

Bond Details

AI Brain Learns

Many 3D Shapes

Results: What Vilya-1 Achieves

Vilya-1 predicts molecule shapes much more accurately than old methods, even for tricky molecules, and can also predict how well they might work as drugs.

Old Shape Predictors

Vilya-1 Predictor

Compare Accuracy

Vilya-1: Best Shapes

Vilya-1: Drug Properties

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.