SOLVAR: Fast covariance-based heterogeneity analysis with pose refinement for cryo-EM
New algorithm tackles a key computational bottleneck in structural biology, enabling faster 3D molecular mapping.
Deep Dive
Researchers Roey Yadgar and Yoel Shkolnisky developed SOLVAR, a new machine learning algorithm for cryo-electron microscopy (cryo-EM). It uses a low-rank assumption on the covariance matrix to efficiently model molecular structural variability. Crucially, it also refines particle image poses, a feature absent from other covariance methods. Benchmarks show state-of-the-art performance, and the code is freely available, offering a faster tool for analyzing continuous molecular heterogeneity.
Why It Matters
This accelerates drug discovery and basic research by making high-resolution 3D molecular structure analysis more computationally efficient.