NVIDIA's BioNeMo expands AI capabilities for drug discovery

NVIDIA's BioNeMo is expanding its computer-aided drug discovery capabilities, with new foundation models intended to revolutionise the field. This generative AI toolkit, designed for pharma and biology researchers, can now provide insights into genomics and protein design. The latest models can analyse DNA sequences, predict alterations in protein shapes after a drug molecule introduction, and determine a cell's function based on its RNA. Over 100 companies, including Cadence and Iambic Therapeutics, are reported to be adopting NVIDIA AI technology to propel their drug discovery and generative AI processes.

Effective use of these models can result in generative AI being seamlessly integrated into the workflows of drug discovery teams, thus reducing the need for resource-intensive physical experiments and offering a greater understanding of drug molecule design. These models are accessible through NVIDIA NIM, a recently announced compilation of inference models, and can be accessed via the NVIDIA AI Enterprise platform. Soon, BioNeMo models will also be accessible via AWS HealthOmics, a service purpose-built to help healthcare and life sciences organisations to store, query, and analyse biological data.

Among the latest foundation models offered by BioNeMo is DNABERT, its first genomics model. This model is trained on DNA sequences and can be utilised to predict the function of specific genome regions, analyse gene mutation and variant effects, and more. Another model, scBERT, trained on single-cell RNA sequencing data, will soon be added to BioNeMo's portfolio. This model supports a variety of applications, such as predicting gene knockout effects, where a specific gene is removed or deactivated, or identifying different cell types. Furthermore, a third model called EquiDock is part of BioNeMo's collection, which can predict the three-dimensional structure of protein interactions, a vital aspect of determining a drug molecule's effectiveness.

Around two dozen healthcare models are currently available in the NIM catalogue of containerised AI microservices. Notables include DiffDock, wich predicts the three-dimensional structures of potential drug candidates and their protein counterparts, and ESMFold, capable of predicting protein structure based on a single amino acid sequence. Another NIM, MolMIM, is able to generate drug candidates optimised for user-defined properties, even designing molecules specifically optimised to bind with a certain protein target. Developed by NVIDIA-Certified Systems, these production-grade NIM microservices can be accessed through NVIDIA AI Enterprise, as well as leading cloud marketplaces.

BioNeMo-powered AI is integrated into the drug discovery workflows of over 100 companies. Tokyo's Astellas Pharma uses BioNeMo to accelerate molecular simulations and language models in drug discovery, utilizing the Tokyo-1 AI supercomputer for further advancements. Meanwhile, Cadence, a computation software developer, is integrating BioNeMo microservices with its Orion platform to increase molecular simulation speed.

San Diego based drug discovery company Iambic has adopted BioNeMo, and will contribute its NeuralPLexer model as a BioNeMo cloud API to assist researchers in predicting how a drug molecule will alter a protein's three-dimensional structure. Insilico Medicine of New York City has also integrated BioNeMo into its workflow and has developed a pipeline of over 30 therapeutic assets. Other companies making use of BioNeMo include Salt Lake City-based Recursion and Southern California biotech firm Terray Therapeutics.