Drug discoverers are among the customers Nvidia has identified in its recently announced catalog of pre-trained, customizable workflows that enable enterprise-level users to create their own artificial intelligence (AI) applications. Masu.
NVIDIA NIM™ Agent Blueprints are reference workflows designed to help drug discoverers and other customers build and deploy generative AI applications for applications such as virtual screening, information retrieval, and even customer service avatars.
Blueprints include NVIDIA NIM or cloud-optimized software designed to help developers accelerate the deployment of generated AI models anywhere, including local workstations, on-premises data centers, cloud services, and GPU-accelerated workstations. Built-in native “microservices”. (NIM stands for “NVIDIA Inference Microservices”).
NIM Agent Blueprint helps researchers design better molecules faster by enabling biopharmaceuticals to move from traditional fixed-database screening to generative AI-driven molecular design and pre-optimization Designed. According to Nvidia, this represents a paradigm shift in the drug discovery process, particularly the move from “hit” compounds to “lead” compounds that are optimized for further development.
“Many of the pharmaceutical companies I go to have 60 million molecules in their library. And that’s what they’re screening, that static 60 million molecules. But… There are 1,060 molecules in chemistry that have the potential to be therapeutics,” Kimberly Powell, vice president of Nvidia Healthcare, told GEN Edge.
“The paradigm shift here is what generative AI is doing, and specifically what the MolMIM system is doing: using generative effects to intelligently search the chemical space, perhaps the most synthetic “You can make molecules that have never been synthesized have characteristics that you’ve never synthesized before,” Powell said.
For drug developers, NIM Agent Blueprint, called Generative Virtual Screening, reduces the time and cost of developing new therapies by using generative models to accelerate virtual screening of small molecules, according to Nvidia. It will help realize the long-held promise of AI.
Improve “hit” with three AI models
Blueprints identify and improve virtual “hit” compounds identified through screening as having potential biological activity in a smarter and more efficient way. At the core of generative virtual screening are three key AI models:
AlphaFold2 is an AI model of protein folding developed by Google DeepMind. AlphaFold2 can predict the 3D structure of proteins from amino acid sequences with atomic-level accuracy. DiffDock is a molecular docking model designed to predict the binding structure of small molecule ligands to proteins while simultaneously optimizing multiple properties such as high solubility and low toxicity. MolMIM is a generative chemistry model that generates drug candidates optimized for user-defined properties. MolMIM can also design molecules optimized to bind to specific protein targets.
Each AI model is packaged within NIM, which integrates microservices into flexible, scalable generative AI workflows. Blueprint uses a generative AI approach to pre-optimize molecules for desired therapeutic properties.
“Virtual screening is still only one part of drug discovery, but we are working on models that span from target discovery to lead identification, and we are building blueprints along that drug discovery process. We’re going to keep building,” Powell said.
“Computer-aided drug discovery really introduces generative AI in the process,” she explained. “Computer-aided drug discovery is often thought of as optimizing lead identification, and we do a lot of simulation there. method.”
Other NVIDIA NIMs focused on drug discovery include:
ESMFold is a “transformer” model, a neural network that learns context and meaning by tracking relationships in continuous data, that can accurately predict protein structure based on a single amino acid sequence. Parabricks DeepVariant (the tool behind the Universal Variant Calling Microservice) is a deep learning model designed to help identify variants in short-read and long-read sequence datasets. Parabricks is designed to improve the speed of variant calling in genomic analysis workflows by a factor of 50 compared to the original or “vanilla” DeepVariant implementation, which was designed to run on a central processing unit or CPU.
“There are up to four to five NIMs generally available, and many more drug discovery and healthcare NIMs in preview. We intend to be active in announcing and deploying these applications. “Every month, we’ll have a wealth of new NIMs and blueprints,” Powell said.
Digital human, PDF data extraction
In addition to the drug discovery blueprint, other NIM agent blueprints include digital human workflows for applications ranging from digital health to customer service. The multimodal PDF data extraction workflow for enterprise search augmented generation (RAG) is designed to generate more accurate responses from vast amounts of business data.
According to Nvidia, RAG can read images from any PDF and provide insights based on what it sees.
“In the healthcare industry, whether it’s biomedical research, all the interactions with insurance companies, or patient-doctor interactions, there are PDFs everywhere that contain tons of useful information. We’re going to extract and summarize it. We will be able to do it,” Powell said.
Digital human workflows can also be applied to digital health, using customized avatars (such as Nvidia’s interactive digital human named “James”) capable of automatic speech recognition. Human speech is converted to text, enters the language model, and from there is sent to the RAG system and back to speech synthesis to activate the user’s avatar.
“When you run that complete loop, you have a true digital human representation that can understand, reason, and respond, and also use Audio2Face NIM,” Powell said. “When you meet James and the text-to-speech comes back, you actually see a different emotion on your face and it makes for a more engaging conversation.”
Blueprints are free for developers to download and deploy on the NVIDIA AI Enterprise software platform.
According to Nvidia, biopharmaceuticals, including all 20 of the top companies in the field, have access to NVIDIA NIM Agent Blueprints through global system integrators and technology solution providers such as Accenture, Deloitte, Softserve, and World Wide Technology (WWT) to Provided to companies in the middle. Cisco, Dell Technologies, Hewlett Packard Enterprise, and Lenovo offer full-stack NVIDIA acceleration infrastructure and solutions to accelerate the deployment of NIM agent blueprints.
Accenture plans to customize NIM agent blueprints to the specific needs of drug development programs by partnering with biopharmaceuticals to optimize the molecule generation steps within MolMIM NIM.
Amazon Web Services’ AWS HealthOmics, a service designed to help biopharmaceutical and healthcare systems store, query, and analyze genomic, transcriptomic, and other omics data, uses Blueprints to streamline integration. We have all three NIMs available for configuration. Nvidia said it will introduce AI into existing drug discovery workflows.
