NVIDIA Pioneers AI Solutions in Drug Discovery with New Customizable Workflows
Santa Clara, CA—NVIDIA has announced a suite of pretrained, customizable artificial intelligence (AI) workflows, aimed at revolutionising various industries, including the pharmaceutical sector. Titled NVIDIA NIM™ Agent Blueprints, these reference workflows are designed to assist enterprises in building and deploying generative AI applications efficiently. Notable applications include virtual screenings, information retrieval, and customer service avatars.
NIM Agent Blueprints use AI models across diverse computing environments, from local workstations to cloud services. These Blueprints incorporate NVIDIA Inference Microservices (NIMs), which optimise cloud-native microservices to fast-track the deployment of generative AI models. The initiative is set to transform how biopharma companies conduct molecule screening, shifting from fixed database screening to generative AI-driven design and pre-optimization.
Kimberly Powell, NVIDIA’s business development manager for healthcare, highlighted the vast potential unlocked by this innovation. “Many drug companies traditionally screen against a static library of about 60 million molecules. However, the potential chemical space contains 10^60 molecules. Our generative AI helps intelligently search this space to find new, potentially therapeutic molecules that have never been synthesized,” Powell explained.
NVIDIA's Blueprint particularly targets drug developers with its generative virtual screening capabilities. This process aims to speed up and reduce the cost of drug development by leveraging three key AI models:
AlphaFold2: Developed by Google DeepMind, this model predicts 3D structures of proteins from amino acid sequences with atomic-level accuracy.
DiffDock: A molecular docking model that predicts the binding structure of small molecule ligands to proteins, optimising multiple properties such as solubility and toxicity.
MolMIM: This generative chemistry model creates drug candidates based on user-defined properties, including optimisation for binding to specific protein targets.
These AI models are encapsulated within NIMs, integrated into a flexible and scalable generative AI workflow. The pre-optimisation of molecules for therapeutic properties represents a significant advancement in drug discovery, moving beyond mere lead identification towards a comprehensive computational approach from target discovery to lead optimisation.
NVIDIA also offers other NIMs tailored to drug discovery, including:
- ESMFold: A neural network (Transformer model) for accurate protein structure prediction from a single amino acid sequence.
- Parabricks DeepVariant: A deep learning model enhancing variant calling in genomic analysis, boasting a 50x speed improvement over previous implementations.
NVIDIA’s strategic partners for deploying NIM Agent Blueprints include top biopharma companies, global system integrators like Accenture, Deloitte, SoftServe, and World Wide Technology (WWT). Hardware partners like Cisco, Dell Technologies, Hewlett Packard Enterprise, and Lenovo are also incorporating NVIDIA-accelerated infrastructure to facilitate these deployments.
Accenture, for example, is working to customise a NIM Agent Blueprint to meet specific needs in drug development, particularly enhancing the molecule generation phase within MolMIM. Moreover, Amazon Web Services (AWS) is integrating these workflows within its HealthOmics service, aiming to streamline AI integration in existing drug discovery processes.
NVIDIA’s initiative extends beyond drug discovery, with Blueprints for digital human workflows and multimodal PDF data extraction. Their digital human workflow creates interactive avatars for applications in digital health and customer service. This integrates automatic speech recognition with a language model, converting human speech into text and generating responsive speech synthesis for avatars.
Ultimately, NVIDIA’s NIM Agent Blueprints provide a versatile and innovative solution for multiple industries. Available for free download and compatible with NVIDIA’s AI Enterprise software platform, these Blueprints are poised to facilitate the creation of sophisticated AI applications across a wide array of domains.
Source: Noah Wire Services