Particle quality assessment and sorting for automatic and semiautomatic particle-picking techniques

Three-dimensional reconstruction of biological specimens using electron microscopy by single particle methodologies requires the identification and extraction of the imaged particles from the acquired micrographs. Automatic and semiautomatic particle selection approaches can localize these particles, minimizing the user interaction, but at the cost of selecting a non-negligible number of incorrect particles, which can corrupt the final three-dimensional reconstruction. In this work, we present a novel particle quality assessment and sorting method that can separate most erroneously picked particles from correct ones. The proposed method is based on multivariate statistical analysis of a particle set that has been picked previously using any automatic or manual approach. The new method uses different sets of particle descriptors, which are morphology-based, histogram-based and signal to noise analysis based. We have tested our proposed algorithm with experimental data obtaining very satisfactory results. The algorithm is freely available as a part of the Xmipp 3.0 package [http://xmipp.cnb.csic.es].