2026Chen MPM
Citation
Chen, J., Leung, V.C., Wang, R., Bubeck, D. and Dragotti, P.L. 2026. Masked Projection Modelling for Sparse-view cryo-EM Reconstruction. ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (2026), 11567–11571.
Abstract
Resolving conformational heterogeneity in cryo-electron microscopy (cryo-EM) remains challenging, especially for rare states. Standard reconstruction methods, reliant on abundant similar particles, bias results toward dominant conformations. To address this, we present an end-to-end pipeline that separates orientation estimation from 3D reconstruction. Our approach starts with self-supervised Masked Projection Modelling (MPM), which pretrains an encoder to capture geometric relationships across projections without known orientations. This encoder drives a supervised Probabilistic Orientation Estimation (POE) framework for initial orientation inference. At testing stage on real data, a high-resolution 3D volume is estimated while orientations are further refined within a Maximum-Likelihood Expectation-Maximization (ML-EM) algorithm that utilizes a continuous Implicit Neural Representation (INR), requiring no pretraining data. On real cryo-EM data, our method achieves high-resolution reconstruction from a severely reduced number of particle projections, outperforming traditional methods in low-particlecount scenarios.
Keywords
Links
https://ieeexplore.ieee.org/abstract/document/11463226/