Ab initio cryo-EM structure determination as a validation problem

Abstract

Robust ab initio structure determination and outcome validation are significant challenges that limit the potential of single particle structural EM. Stochastic Hill Climbing (SHC) can improve the convergence properties of widely-used projection matching (PM) algorithms and facilitate ab initio structure determination, but outcome correctness remains in question as SHC is an local optimizer. We propose a protocol, VIPER, for determination and validation of ab initio structures based on an evolutionary approach. The outcomes of parallel independent runs are brought into register, monitored for consistency, and crossovers are conducted to improve overall convergence behavior. After each generation, the best candidates are selected according to their L2 norm values used as a goal function. Screening based on reproducibility of individual image parameters allows for outlier elimination. This new approach introduces rigorous outcome validation into EM ab initio structure determination.