Youdong Jack Mao
Harvard Medical School, USA
Title: Advanced machine-learning algorithms for improving high-performance cryo-EM data analysis
Biography
Biography: Youdong Jack Mao
Abstract
Machine learning technology represents an intriguing avenue in the methodology development for cryo-EM structure determination. In this presentation, we explore two aspects of machine learning development that help advance cryo-EM data analysis. First, structural heterogeneity in single-particle images presents a major challenge for high-resolution cryo-EM structure determination. Recently we introduce a statistical manifold learning approach for unsupervised single-particle deep classification. When optimized for Intel High-Performance Computing (HPC) processors, our approach implemented in ROME software package, can generate thousands of reference-free class averages within several hours from hundreds of thousands of single-particle cryo-EM images. Deep classification thus assists in computational purification of single-particle datasets for high-resolution reconstruction. Second, particle extraction represents a major practical bottleneck in the structure determination of biological macromolecular complexes by single-particle cryo-EM. We developed a deep learning-based algorithmic framework, DeepEM, for single-particle recognition from noisy cryo-EM micrographs, enabling automated particle picking, selection and verification in an integrated fashion. Our approach exhibits improved performance and high accuracy when tested on the standard KLH dataset as well as several challenging experimental cryo-EM datasets.
