Ab initio CryoEM maps minimizing preferred orientation problems in heterogeneous mixtures: ReconSIREN
- 1National Centre for Biotechnology (CSIC)
Abstract
Cryo-Electron Microscopy (CryoEM) Single Particle Analysis (SPA) is pivotal for elucidating protein structures and their conformational dynamics. Several critical steps are required to obtain accurate maps. First, the \textit{ab initio} relative orientations and shifts between images must be determined. Second, the common issue of specimens presenting a preferred orientation should be adequately addressed. Third, all of the above must account for particle flexibility and sample heterogeneity. Additionally, these steps are interrelated, making the search for all these parameters a real challenge when considering convergence robustness. In this context, we present ReconSIREN, a novel deep learning framework that addresses the key steps outlined before in an \textit{ab initio} manner while greatly minimizing one of the most important practical burdens in SPA, namely the errors introduced in the CryoEM maps by the existence of preferred orientations; the latter is achieved by a change of the optimization approach that avoids the so-called ``attraction problem'', and by the introduction of a new type of geometrical priors. It does so fully \textit{ab initio}, using deep learning training/inference steps without resorting to more traditional optimization approaches, and generating highly accurate initial reconstructions. We present maps as both homogeneous (consensus) maps and a full conformational landscape for analyzing flexibility and heterogeneity. Thanks to its Mixture-of-Experts (MoE) architecture, ReconSIREN distributes the estimation workload across independent experts, improving optimization efficiency and simplifying the process.
Related
- Positive Selection Drives Preferred Segment Combinations during Influenza Virus Reassortment
- Preferred visual experiences provide intuitive descriptions of the functional properties of the cortical navigation network
- Association between preferred and documented mode of birth and birth experiences in low- and middle-income countries: a multisite cross-sectional study
- The development of a technologic approach to improve access to individualized clinical documentation in caregivers’ preferred language
- Examining microbial assemblage differences on plants preferred and avoided by threatened Hawaiian land snails to identify important food resources
BibTeX
@misc{herreros2026initio,
title = {{Ab initio CryoEM maps minimizing preferred orientation problems in heterogeneous mixtures: ReconSIREN}},
author = {D. Herreros and C. O. S. Sorzano and J. M. Carazo},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.03.756391},
url = {https://www.biorxiv.org/content/10.64898/2026.10.03.756391v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Ab initio CryoEM maps minimizing preferred orientation problems in heterogeneous mixtures: ReconSIREN
AU - D. Herreros
AU - C. O. S. Sorzano
AU - J. M. Carazo
PY - 2026
DA - 2026/10/03
PB - bioRxiv
DO - 10.64898/2026.10.03.756391
UR - https://www.biorxiv.org/content/10.64898/2026.10.03.756391v1
AB - Cryo-Electron Microscopy (CryoEM) Single Particle Analysis (SPA) is pivotal for elucidating protein structures and their conformational dynamics. Several critical steps are required to obtain accurate maps. First, the \textit{ab initio} relative orientations and shifts between images must be determined. Second, the common issue of specimens presenting a preferred orientation should be adequately addressed. Third, all of the above must account for particle flexibility and sample heterogeneity. Additionally, these steps are interrelated, making the search for all these parameters a real challenge when considering convergence robustness. In this context, we present ReconSIREN, a novel deep learning framework that addresses the key steps outlined before in an \textit{ab initio} manner while greatly minimizing one of the most important practical burdens in SPA, namely the errors introduced in the CryoEM maps by the existence of preferred orientations; the latter is achieved by a change of the optimization approach that avoids the so-called ``attraction problem'', and by the introduction of a new type of geometrical priors. It does so fully \textit{ab initio}, using deep learning training/inference steps without resorting to more traditional optimization approaches, and generating highly accurate initial reconstructions. We present maps as both homogeneous (consensus) maps and a full conformational landscape for analyzing flexibility and heterogeneity. Thanks to its Mixture-of-Experts (MoE) architecture, ReconSIREN distributes the estimation workload across independent experts, improving optimization efficiency and simplifying the process.
ER -