Nvidia BioNeMo Models
With the integration of NVIDIA NIM, a set of cloud-native microservices, with Amazon Web Services, utilising optimised AI models for healthcare is now simpler than ever.
Through industry-standard application programming interfaces, or APIs, NIM, a component of the NVIDIA AI Enterprise software platform offered on the AWS Marketplace, gives developers access to an expanding library of AI models. With enterprise-grade security and support, the library offers foundation models for drug discovery, medical imaging, and genomics.
NIM may now be accessed through AWS ParallelCluster, an open-source platform for managing and deploying high performance computing clusters on AWS, and Amazon SageMaker, a fully managed service for preparing data and building, training, and deploying machine learning models. Another tool for orchestrating NIMs is AWS HealthOmics, a service designed specifically for biological data processing.
The hundreds of healthcare and life sciences businesses that currently use AWS will be able to implement generative AI more quickly thanks to easy access to NIM, eliminating the hassles associated with model building and production packaging. Additionally, it will assist developers in creating workflows that integrate AI models with data from many modalities, including MRI scans, amino acid sequences, and plain-text patient health records.
This initiative, which was presented today at the AWS Life Sciences Leader Symposium in Boston, expands the range of NVIDIA Clara accelerated healthcare software and services that are available on AWS. These services include NVIDIA BioNeMo‘s quick and simple-to-deploy NIMs for drug discovery, NVIDIA MONAI for medical imaging workflows, and NVIDIA Parabricks for accelerated genomics.
Pharmaceutical and Biotech Businesses Use NVIDIA AI on Amazon
Nvidia BioNeMo is a generative AI platform that supports the training and optimisation of biology and chemistry models on private data. It consists of foundation models, training frameworks, domain-specific data loaders, and optimised training recipes. Over a hundred organisations utilise it worldwide.
One of the top biotechnology firms in the world, Amgen, has trained generative models for protein design using the Nvidia BioNeMo framework and is investigating the possibility of integrating Nvidia BioNeMo with AWS.
The Nvidia BioNeMo models for molecular docking, generative chemistry, and protein structure prediction are pretrained and optimised to run on any NVIDIA GPU or cluster of GPUs. They are available as NIM microservices. Combining these models can enable a comprehensive approach for AI-accelerated drug discovery.
A-Alpha Bio is a biotechnology business that uses artificial intelligence (AI) and synthetic biology to quantify, forecast, and design protein-to-protein interactions. Researchers witnessed a speedup of more than 10x as soon as they switched from a generic version of the ESM-2 protein language model to one that was optimised by NVIDIA and ran on NVIDIA H100 Tensor Core GPUs on AWS. As a result, the team is able to sample a far wider range of protein possibilities than they otherwise could have.
Using retrieval-augmented generation, or RAG, also referred to as a lab-in-the-loop architecture, NIM enables developers to improve a model for organisations who wish to supplement these models with their own experimental data.
Accelerated Genomics Pipelines Made Possible by Parabricks
NVIDIA Parabricks genomics models are included in NVIDIA NIM and can be accessed on AWS HealthOmics as Ready2Run workflows, which let users set up pre-made pipelines.
The life sciences company Agilent greatly increased the processing rates for variant calling workflows on its cloud-native Alissa Reporter software by utilising Parabricks genomics analysis tools running on NVIDIA GPU-powered Amazon Elastic Compute Cloud (EC2) instances. Researchers can get quick data analysis in a secure cloud environment by integrating Parabricks with Alissa secondary analysis workflows.
Artificial Conversational Intelligence Promotes Digital Health
NIM microservices provide optimised big language models for conversational AI and visual generative AI models for avatars and digital humans, in addition to models that can read proteins and genetic sequences.
By providing logistical support to clinicians and responding to patient inquiries, AI-powered digital assistants can improve healthcare. After receiving training on RAG-specific data from healthcare organisations, they were able to link to pertinent internal data sources to aggregate research, reveal patterns, and boost efficiency.
startup using generative AI AI-powered healthcare agents that concentrate on a variety of tasks like wellness coaching, preoperative outreach, and post-discharge follow-up are now being tested by Hippocratic AI.
The company is implementing Nvidia BioNeMo Models and NVIDIA ACE microservices to power a generative AI agent for digital health. The company employs NVIDIA GPUs through AWS.
The team powered the discussion of an avatar healthcare assistant with NVIDIA Audio2Face facial animation technology, NVIDIA Riva automated voice recognition, text-to-speech capabilities, and more.
NVIDIA created a collection of tools called Nvidia BioNeMo models especially for use in life sciences research, including drug development. They are constructed around the Nemo Megatron framework from NVIDIA, which is a toolkit for creating and honing massive language models.
Features of Nvidia BioNeMo Models
Pre-conditioned AI models
Large volumes of biological data have already been used to train these models. Then, these models can be applied to a range of activities, including determining possible drug targets, assessing the impact of mutations, and forecasting protein function. Pre-trained Nvidia BioNeMo models include, for instance-
DNABERT:
This model is useful for analysing and forecasting the function of DNA sequences.
ScBERT:
This model can be used to identify distinct cell types and forecast the consequences of gene knockouts because it was developed using single-cell RNA sequencing data.
EquiDock:
The 3D structure of protein interactions may be predicted using this approach, which is useful for finding possible therapeutic options.
BioNeMo Service
Researchers can simply access and utilise Nvidia BioNeMo‘s pre-trained models through a web interface by using the BioNeMo Service, a cloud-based solution . For researchers without access to the computational power needed to train their own models, this service can be especially helpful.
All things considered, Nvidia BioNeMo models are an effective instrument that can be utilised to quicken medication discovery research. These models help researchers find novel drug targets and create new treatments more swiftly and effectively.
