PartiNet is a dynamic adaptive neural network for high-performance particle picking in cryo-electron microscopy

Abstract

Accurate, efficient and autonomous particle picking is a major bottleneck in high-resolution cryo-electron microscopy (cryo-EM). We introduce PartiNet, an Artificial Intelligence (AI)-based particle picker with size-agnostic detection and pre-trained models that eliminate manual parameter specification and dataset-specific training, pioneering dynamic neural network inference for single particle cryo-EM pipeline. Unlike static architectures, PartiNet employs a dynamic framework that adjusts network complexity in real-time based on perceived micrograph quality. This adaptive approach accelerates inference up to 7-fold compared to existing tools without sacrificing particle selection quality. Training on diverse protein datasets showed that PartiNet improves particle yields, enhances sampling of rare orientations, and is compatible with on-the-fly workflows. Comprehensive evaluation on benchmark datasets and validation on a new dataset of the chromatin remodeler MORC2 demonstrates superior precision and recall, with the ability to detect heterogeneous protein species, leading to more complete structural models and consistently higher-resolution reconstructions.