Cryo-electron microscopy (cryo-EM) represents a powerful technology for determining atomic models of biological macromolecules. Despite this promise, human-guided cryo-EM data collection practices limit the impact of cryo-EM because of a path planning problem: cryo-EM datasets typically represent 2-5% of the total sample area. Here, we address this fundamental problem by formalizing cryo-EM data collection as a path planning optimization from low signal data. Within this framework, we incorporate reinforcement learning (RL) and deep regression to design an algorithm that uses distributed surveying of cryo-EM samples at low magnification to learn optimal cryo-EM data collection policies. Our algorithm - cryoRL - solves the problem of path planning on cryo-EM grids, allowing the algorithm to maximize data quality in a limited time without human intervention. A head-to-head comparison of cryoRL versus human subjects shows that cryoRL performs in the top 10% of test subjects, surpassing the majority of users in collecting high-quality images from the same sample. CryoRL establishes a general framework that will enable human-free cryo-EM data collection to increase the impact of cryo-EM across life sciences research.

Compositional AI systems, which combine multiple artificial intelligence components together with other application components to solve a larger problem, have no known pattern of development and are often approached in a bespoke and ad hoc style. This makes development slower and harder to reuse for future applications. To support the full rapid development cycle of compositional AI applications, we have developed a novel framework called (Bee) * (written as a regular expression and pronounced as "beestar"). We illustrate how (Bee) * supports building integrated, scalable, and interactive compositional AI applications with a simplified developer experience.

Cryo-electron microscopy (cryo-EM) represents a powerful technology for determining atomic models of biological macromolecules(Kühlbrandt, 2014). Despite this promise, human-guided cryo-EM data collection practices limit the impact of cryo-EM because of a path planning problem: cryo-EM datasets typically represent 2-5% of the total sample area. Here, we address this fundamental problem by formalizing cryo-EM data collection as a path planning optimization from low signal data. Within this framework, we incorporate reinforcement learning (RL) and deep regression to design an algorithm that uses distributed surveying of cryo-EM samples at low magnification to learn optimal cryo-EM data collection policies. Our algorithm - cryoRL - solves the problem of path planning on cryo-EM grids, allowing the algorithm to maximize data quality in a limited time without human intervention. A head-to-head comparison of cryoRL versus human subjects shows that cryoRL performs in the top 10% of test subjects, surpassing the majority of users in collecting high-quality images from the same sample. CryoRL establishes a general framework that will enable human-free cryo-EM data collection to increase the impact of cryo-EM across life sciences research.

Cryo-electron microscopy (cryo-EM) represents a powerful technology for determining atomic models of biological macromolecules(Kühlbrandt, 2014). Despite this promise, human-guided cryo-EM data collection practices limit the impact of cryo-EM because of a path planning problem: cryo-EM datasets typically represent 2-5% of the total sample area. Here, we address this fundamental problem by formalizing cryo-EM data collection as a path planning optimization from low signal data. Within this framework, we incorporate reinforcement learning (RL) and deep regression to design an algorithm that uses distributed surveying of cryo-EM samples at low magnification to learn optimal cryo-EM data collection policies. Our algorithm - cryoRL - solves the problem of path planning on cryo-EM grids, allowing the algorithm to maximize data quality in a limited time without human intervention. A head-to-head comparison of cryoRL versus human subjects shows that cryoRL performs in the top 10% of test subjects, surpassing the majority of users in collecting high-quality images from the same sample. CryoRL establishes a general framework that will enable human-free cryo-EM data collection to increase the impact of cryo-EM across life sciences research.