Medicine

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

Pascal Sturmfels, Milad Salem, Naozumi Hiranuma, Stephen Rettie, Xiaoliang Pan, Benjamin D. Sellers, Adam P. Moyer, Patrick J. Salveson, Ivan Anishchanka

Featured July 20, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content.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

Simply

By using simple 'detectors' on the hidden patterns learned by a powerful DNA AI model, scientists can quickly spot dangerous genes like those causing antibiotic resistance in mixed genetic samples, even from tiny fragments.

In depth

The paper explores the use of genomic foundation model embeddings, specifically from Evo 2, for rapid biosecurity screening. They demonstrate that lightweight linear and attention probes, trained on frozen intermediate activations, can accurately detect antimicrobial resistance (AMR) and bacterial virulence in metagenomic sequences. This approach offers a fast and inexpensive first-pass detection layer without requiring computationally intensive fine-tuning of the large foundation model.

Key Takeaways

1
Lightweight linear and attention probes on frozen Evo 2 embeddings achieve strong discrimination for antimicrobial resistance (AMR) and bacterial virulence in metagenomic data.
2
The attention probe significantly outperforms the linear probe for AMR detection, suggesting that learned token weighting is crucial for identifying relevant genomic signals.
3
Probes trained on long sequences maintain comparable performance on simulated short reads without retraining, highlighting their potential for real-time metagenomic biosurveillance in fragmented data.

Conceptual Flow

HIGH LEVEL

Methodology: Using AI's Hidden Knowledge

The paper uses a big AI model that understands DNA, then adds a tiny 'detector' to its internal thoughts to find specific bad genes without changing the big AI.

Raw DNA Sequence

AI Learns Patterns

Hidden AI Thoughts

Simple Detector

Is Bad Gene Present?

Results: Fast & Accurate Threat Detection

They found that these simple detectors are really good at finding antibiotic resistance and other threats, even in messy, broken-up DNA samples, making it a quick first check.

Mixed DNA Samples

Quickly Scan for Threats

Antibiotic Resistance Found

Virulence Factors Found

Fast First Check