Improving Cryo-EM Optimization Robustness with an Optimal Transport Loss Function for Noisy Images

The Sliced Wasserstein loss provides a smoother optimization landscapes than mean squared error for single particle cryo-EM joint inference of pose, CTF defocus and conformational heterogeneity. Estimating background contrast is essential to avoid biasing other parameters.

Recommended citation: Geoffrey Woollard, David Herreros, Minhuan Li, Pilar Cossio, Khanh Dao Duc. (2025). "Improving Cryo-EM Optimization Robustness with an Optimal Transport Loss Function for Noisy Images." bioRxiv.
Download Paper