Custom Drug Discovery Workflows | Tutorial on NVIDIA BioNeMo Service

Key Takeaways

foreign email service accelerates end-to-end AI drug Discovery pipelines with instant seamless access to generative Ai and large language models an ai-powered cloud-based service it provides pre-trained models with billions of parameters at supercomputing scale these models include Mega mobart and moflow for generative chemistry esm1 and esm2 for protein learned embeddings and property prediction diff dock for pose prediction Pro gpt2 for protein generation and open fold Alpha fold 2 and esm fold for 3D protein structure prediction to get started let's visit the bionimo members area this site includes documentation on the key features of bionemo and step-by-step instructions to get started with the service all models in bionemo are accessible on NGC in the biomal playground UI or with the bionimo service API the first step is logging into NGC where we can find the full NGC catalog the bionemo UI and tools to generate a unique NGC API key to access the API the bionemo UI allows researchers to quickly test model inference with the click of a button using either built-in example queries or custom inputs foreign the results of the inference requests can be visualized directly in the UI or downloaded for use in local Downstream tasks the UI also provides tool tips that demonstrate the use of the bionemo service API through either HTTP requests using curl or with the bionemo python client interface NGC provides a unique API key that allows us to programmatically use NGC models and services like bio Nemo to generate the key navigate to the NGC setup page we'll save a copy of the key since it's our unique identifier for accessing the service we can use an example from the documentation to test that the API key can connect to and authenticate bionemo service to display available models bionemo also includes a set of Jupiter notebooks that demonstrate its python client the python client wraps the bio Nemo rest API to provide simplified one-line calls to all available models the client can be used to generate blocking and non-blocking calls into bionemo service making it possible to build complex workflows over a broad range of models and inputs and the Jupiter notebooks offer step-by-step tours of example workflows that can be used to integrate bionemo into custom applications these workflows include a virtual drug screening pipeline protein structure prediction protein and small molecule generation molecular docking and downstream tasks using learned embeddings Nvidia bionemo Services can be used to quickly test available models and adapt them to custom workflows and Drug Discovery Pipelines visit nvidia.com to learn more about bio Nemo and to find more in-depth examples and tutorials