Revolutionize Drug Discovery: Unleash the Power of 9 Cutting-Edge AI Models for Breakthrough Innovations with BioNeMo

Build Generative AI Pipelines for Drug Discovery with NVIDIA’s Service

NVIDIA BioNeMo Service is a cloud service for generative AI in early drug discovery, featuring nine state-of-the-art large language and diffusion models.

These models are designed to optimize workflows and allow researchers to focus on adapting AI models for drug candidates, rather than dealing with infrastructure. The service can generate large libraries of proteins, refine protein libraries, generate small molecules with specific properties, predict and visualize 3D protein structures, and run campaigns of ligand-to-small-molecule pose estimations.

Non-Biotech Startups Have Access to Biotechnology Can-Do Like Never Before

1. AI models are increasingly used in drug discovery, with about 160 discovery programs and >14 products in clinical development.
2. Generative AI models can predict the 3D structures of proteins, create de novo proteins and molecules, and predict small molecule binding structures.
3. Protein-protein interactions are now being engineered for studies, setting the stage for clinical applications in the near future, if not by the time you’re reading this
4. NVIDIA BioNeMo Service offers nine AI generative models, covering a wide range of applications for AI drug discovery pipelines, including
   • AlphaFold 2,
   • ESMFold,
   • OpenFold,
   • ESM-1nv,
   • ESM-2,
   • ProtGPT2,
   • MegaMolBART,
   • MoFlow, and
   • DiffDock.
5. The models can be accessed through a web interface or APIs, and can be further trained and optimized on NVIDIA DGX Cloud.
6. NVIDIA GTC 2023 conference features several sessions on AI drug discovery and BioNeMo.

Sorting Out When to Use Which

AlphaFold 2

High-accuracy 3D protein structure prediction from amino acid sequences

1. AlphaFold 2: Developed by DeepMind, AlphaFold 2 is a 3D protein structure prediction model that predicts the relationship between amino acid sequences of a protein and its 3D structures with high accuracy, even with few homologous sequences available. It achieved near experimental accuracy for predicted protein 3D structures at CASP14.

ESMFold

Ultrafast, transformer-based, single-sequence 3D protein structure prediction

• ESMFold: Developed by Meta, ESMFold is a transformer-based, ultrafast 3D protein structure prediction model based on ESM-2 embeddings without multiple sequence alignment (MSA). It includes a folding head for a fully end-to-end, single-sequence structure predictor and can predict the structure of a single protein sequence without requiring many homologous sequences as input.

OpenFold

PyTorch-based reproduction of AlphaFold 2 for 3D protein structure prediction.

• OpenFold: OpenFold is a faithful reproduction of DeepMind’s AlphaFold 2 model, but based on PyTorch instead of JAX. It predicts 3D protein structure from a primary amino acid sequence and achieves similar accuracy to the original model.

ESM-1nv

Evolutionary-scale protein modeling based on BERT architecture for protein structure and function prediction.

• ESM-1nv: A reproduction of Meta’s ESM-1b, ESM-1nv is a large language model (LLM) for the evolutionary-scale modeling of proteins. It is based on the BERT architecture and trained on millions of protein sequences with a masked language modeling objective, learning the patterns and dependencies between amino acids that ultimately give rise to protein structure and function.

ESM-2

Mutation impact prediction and protein stability analysis using BERT-inspired architecture.

• ESM-2: ESM-2 is an LLM similar to ESM-1nv, but typically used to predict the effects of mutations on protein stability. It is also based on the BERT architecture and trained on millions of protein sequences.

ProtGPT2

De novo protein sequence generation to identify unique structures, properties, and functions

• ProtGPT2: Developed at ISMB and the Universität of Bayreuth, ProtGPT2 is an LLM based on the GPT2 transformer architecture that generates de novo protein sequences to identify unique structures, properties, and functions. It is optimal for generating custom protein sequences when there is limited training data.

MegaMolBART

Transformer-based molecular optimization and generation of small molecules with binding affinities.

• MegaMolBART: A collaboration between AstraZeneca and NVIDIA, MegaMolBART is a large, transformer-based generative chemistry model used for molecular optimization. It relies on SMILES notation for representing chemical structures of small molecules and is optimal for generating new small molecules with binding affinities that have been tested experimentally, as well as for molecule embeddings.

MoFlow

Flow-based generative model for molecule generation, reconstruction, and optimization using graph convolutions.

• MoFlow: Developed by a team at Weill Cornell Medicine, MoFlow is a flow-based generative model that learns invertible mappings between molecular graphs and their latent representations. It is used for molecule generation, reconstruction, and optimization and achieves state-of-the-art performance with a novel, conditional flow-based approach using graph convolutions.

DiffDock

Diffusion generative AI model for predicting molecular docking and binding structure of small molecules to proteins.

• DiffDock: Developed at MIT’s Jameel Clinic, DiffDock is a diffusion generative AI model that predicts the binding structure of a small molecule ligand to a protein, known as molecular docking or pose prediction. It has fast inference times, provides confidence estimates with high selective accuracy, and is highly accurate and computationally efficient.

Generative AI is democratizing drug discovery

Generative AI is indeed revolutionizing the drug discovery process by democratizing access to advanced computational tools and resources. Platforms like Enthereal are empowering small startups and researchers to explore potential drug leads, products, and risk mitigation strategies more efficiently and effectively than ever before.

By providing turn-key solutions and the necessary know-how, Enthereal allows even small teams to harness the power of generative AI, enabling them to make informed decisions and accelerate their drug discovery pipelines. This ultimately benefits the global community by bringing new, innovative therapeutic options to the market faster and at a lower cost, addressing unmet medical needs and improving the overall quality of healthcare.