Detecting and Interpreting Unexpected Artifacts as Functional States in Cryo-EM Reconstructions

While recent tools like qFit (Wankowicz et al., 2024) and CryoDRGN (Hu & Pande, 2024) focus on modeling conformational heterogeneity, they primarily aim to fit ensemble models to explain observed data. However, unusual or poorly fitting density—often dismissed as noise or artifact—may in fact reveal rare, transient, or previously unknown functional states. This project proposes an AI-powered pipeline to scan cryo-EM maps for systematic deviations from expected density (e.g., outlier regions, unexplained blobs), and to classify them using features learned from both simulated artifacts and known biological variability. By cross-referencing with biochemical data and molecular dynamics simulations, this approach could turn "artifacts" into biological discoveries, leading to a richer understanding of biomolecular dynamics. Unlike conventional model-building which filters out such deviations, this method embraces them as sources of functional insight.

References:

1. Automated multiconformer model building for X-ray crystallography and cryo-EM. Stephanie A. Wankowicz, Ashraya Ravikumar, Shivani Sharma, Blake T. Riley, Akshay Raju, Daniel W Hogan, Jessica Flowers, Henry van den Bedem, D. Keedy, J. Fraser (2024). eLife.
2. End-to-End Training of Latent Space Diffusion Models for Conformational Heterogeneity in Cryo-EM Reconstruction. Zixi Hu, Kanupriya Pande (2024). Image and Vision Computing New Zealand.