STTR Phase I: CryoDiscovery?: An integrated cryo-EM intelligence solution

The broader impact of this Small Business Technology Transfer (STTR) Phase I project will be to accelerate discoveries of new molecular structures using cryogenic electron microscopy ("Cryo-EM"). Cryo-EM produces high-resolution 3D images at microscopic levels and is used by researchers in many fields including life sciences, materials science, nanotechnology, semiconductors, energy, environmental science, and food science. Microscopy advancements enable molecular image capture at unprecedented levels of resolution, but the data produced are growing exponentially and subsequent processing of those images into visible 3D structures is both challenging and time consuming. Each project can produce more than 100,000 images and take weeks to arrive at one viewable 3D structure. Current image processing and data analysis solutions are not well-integrated, requiring extensive manual user involvement and long wait times before assessing image quality. We will apply machine learning to automate cryo-EM image processing to improve researcher productivity and accuracy. We will also design the system to reduce user training time. The result will improve access to cryo-EM and accelerate new breakthroughs in many areas of science. This Small Business Technology Transfer (STTR) Phase I project automates image processing for single particle analysis by developing new machine learning models that recognize particles with repeatable accuracy levels and integrates them into the cryo-EM workflow for easy deployment. Images generated by cryo-EM are highly noisy, and the goal is to process them to build recognizable 3D molecular structures. Many steps in the cryo-EM workflow require manual intervention and analysis that can take several weeks and result in errors due to user bias, time waiting and user fatigue. The objectives of this research are to produce a prototype that consistently and accurately predicts particles and is easily integrated into the cryo-EM workflow. The approach will be to increase the training and validation datasets from a wide range of applications and utilize existing convolutional neural network frameworks. We will develop new techniques for running experiments to optimize the models, integrate the prototype into established cryo-EM workflows for end-to-end processing, and produce a delivery method for easy deployment. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.