Cryo-RALib – a modular library for accelerating alignment in cryo-EM
With the enhancement of algorithms, cryo-EM has become the most efficient technique to solve structures of molecules. Take a recent event for example, after the outbreak of COVID-19 in January, the first structure of 2019-nCoV Spike trimer was published in March using cryo-EM, which has provided crucial medical insight for developing vaccines. The enabler behind this efficiency is the GPU-accelerated computation which shortens the whole analysis process to 12 days. However, the data characteristics include strong noise, huge dimension, large sample size and high heterogeneity with unknown orientations have made analysis very challenging. Though, the popular GPU-accelerated Bayesian approach has been shown to be successful in 3D refinement. It is noted that the traditional method based on multireference alignment may better differentiate subtle structure differences under low signal to noise ratio (SNR). However, a modular GPU-acceleration package for multireference alignment is still lacking in the field. In this work, a modular GPU-accelerated alignment library called Cryo-RALib is proposed. The library contains both reference-free alignment and multireference alignment that can be widely used to accelerate state-of-the-art classification algorithms. In addition, we connect the cryo-EM image analysis with the python data science stack which enables users to perform data analysis, visualization and inference more easily. Benchmark on the TaiWan Computing Cloud container, our implementation can accelerate the computation by one order of magnitude. The library has been made publicly available at https://github.com/phonchi/Cryo-RAlib
READ FULL TEXT