NEWS – A new deep learning system called BioEmu predicts the full range of shapes a protein naturally explores under biological conditions.

HIGHLIGHTS

• What is BioEmu? A generative deep learning system named Biomolecular Emulator (BioEmu) that predicts the full range of protein shapes under biological conditions.

• Developed by: Microsoft, Rice University (USA), and Freie Universität (Germany).
• Key Capabilities:
  • Models the equilibrium distribution of protein structures from amino acid sequences.
  • Generates thousands of protein conformations per hour using a single GPU.
  • Enables high-resolution protein flexibility modelling at scale.
• Advantages Over Classical Methods:
  • Faster and more cost-efficient than molecular dynamics (MD) simulations.
  • Accurately captures large shape changes, local unfolding, and transient cryptic pockets relevant for drug design.
  • Predicts 83% of large and 70–81% of small structural changes accurately.
• Application Highlights:
  • Predicts multiple forms (open/closed) of enzymes like adenylate kinase.
  • Works on intrinsically disordered proteins (no fixed 3D structure).
  • Helps understand mutation impacts on protein stability.
• Limitations:
  • Cannot model drug molecules, pH/temperature changes, membranes, or cell environments.
  • Lacks uncertainty estimation like AlphaFold.
• Significance:
  BioEmu revolutionizes protein modeling by offering a fast, scalable, and flexible approach to understanding protein dynamics crucial for drug discovery and biotechnology.