AXIS: a Lab-in-the-Loop machine learning approach for automated detection of macromolecular crystals

Area of Science:

• Structural biology
• Biophysics
• Drug discovery

Background:

• Macromolecular crystallography is vital for understanding biological mechanisms and drug design.
• High-throughput pipelines are revolutionizing protein structure determination.
• Manual crystal identification remains a significant bottleneck in automated workflows.

Purpose of the Study:

• To develop an AI-based system for automated crystal detection in macromolecular crystallography.
• To address the limitations of manual image inspection in high-throughput screening.
• To enhance the automation of protein structure determination pipelines.

Main Methods:

• Development of AXIS, an AI Crystal Identification System.
• Utilized the DINOv2 computer vision model and transfer learning.
• Trained on MARCO, the largest available crystallization dataset.
• Integrated a Lab-in-the-Loop approach for iterative learning and adaptation.

Main Results:

• AXIS achieves automated crystal detection comparable to human experts.
• The system operates effectively with both visible and UV light images.
• Lab-in-the-Loop enables efficient adaptation to specific experimental conditions.
• Automated annotation of large image datasets is now feasible.

Conclusions:

• AXIS significantly reduces bottlenecks in macromolecular crystallography.
• The AI system improves accuracy and efficiency in crystal identification.
• Facilitates higher levels of automation crucial for fundamental and translational research.
• Enables widespread application of automated crystal detection.