NVIDIA Open-Sourced BioNeMo to Advance Computational Biology for Global Biopharma and Science

Written by Mireia Cuevas Crespo (Reporter)

On November 18th, the global technology company NVIDIA (CA, U.S.) announced the open-sourcing of its BioNeMo Framework, making it accessible to other international pharmaceutical and biotech industry leaders, as well as academic pioneers.  

NVDIA BioNeMo is a comprehensive framework containing programming tools, libraries, and pre-trained AI models. It is designed to advance computational drug discovery by streamlining the development and customization of biomolecular models. NVIDIA aims to help scientists identify potential drug candidates, optimize molecular structures, and predict how new compounds behave in biological systems.  

Harnessing AI to Accelerate Drug Discovery  

The drug discovery process is extremely difficult, time-consuming, and financially risky. This is partly due to the complexity and unpredictable nature of biological systems, as well as the vast amount of fragmented data that must be integrated and analyzed to make accurate predictions.  

One way researchers can develop therapies in a time-efficient manner is by doing computational modelling and predictions. To do this, databases containing specialized biomolecular models and programming tools are needed.  

To facilitate this process, NVIDIA is open-sourcing BioNeMo, a generative AI platform intended to scale biomolecular research for drug discovery and molecular design.  

According to NVIDIA, BioNeMo puts “its range of optimized, pre-trained biomolecular models and workflows, along with versatile functionalities for building and customizing models, including training and fine-tuning, at the service of global organizations.”  

“The convergence of AI, accelerated computing and expanding datasets offers unprecedented opportunities for the pharmaceutical industry, as evidenced by recent Nobel Prize wins in chemistry. To help unravel the complexities of biological systems, we’ve introduced the open-source BioNeMo Framework, which will enable researchers worldwide to accelerate the development of life-saving treatments.”  

Kimberly Powell, Vice President of Healthcare at NVIDIA.  

NVIDIA stated that the tool is already integrated by more than 200 biopharma companies and startups. Several companies are or have agreed to contribute their expertise in AI, biotechnology, and data science to BioNeMo, included in this list are 

A -Alpha Bio (WA, U.S.) and Argonne National Laboratory (IL, U.S.), among others.  This collaboration aims to enhance the framework, extending its use across broader environments.   

NVIDIA’s announcement represents the second major open-source release in the same week, right after Google DeepMind made the source and model codebase of its leading protein structure prediction system, AlphaFold 3, publicly available. 

BioNeMo’s NIM Microservices 

NVIDIA unveiled a brand-new set of user-friendly, optimized, AI-driven microservices, which is a method of designing software systems, where the applications are broken down into smaller, independent units, each responsible for a specific function.  

NVDIA Inference Microservices (NIM) is a set of cloud-native microservices designed to make deploying AI models easier, faster, and more efficient. NIM microservices are modular, optimized, and easy to deploy, allowing developers to easily execute programs in molecular design, protein structure prediction, or cloud environments. This technology aims to accelerate drug development processes, speeding up the path from inference to insights. 

The newly launched NIM microservices are designed to integrate with leading AI models such as:  

  1. AlphaFold2: Capable of predicting protein structures in almost real-time with a 5x speedup.  
  2. DiffDock 2.0: Based on research from MIT and trained on the gold-standard PLINDER dataset, DiffDock 2.0 enables researchers to predict molecular orientations 6.2 times faster and with 16% greater accuracy.   
  3. RFdiffusion & ProteinMPNN: NIM microservices accelerate the design of novel proteins, facilitating the development of new protein-based therapeutics. 
  4. BioNeMo Blueprints: Designed to assist developers in scaling their AI deployments into enterprise-level production pipelines, BioNeMo Blueprint for virtual screening offers a flexible, user-friendly workflow that utilizes NIM microservices to accelerate small molecule design, reducing both time and costs. 

Furthermore, another groundbreaking innovation lies in the development of Acceleration Libraries. An example of this is cuEquivariance, which has been added to BioNeMo to speed up the mathematical calculations required for DiffDock chemical predictions.  

Potential Drawbacks     

AI-driven databases and AI systems have recently demonstrated considerable potential in advancing healthcare, including areas such as cancer, fetal health forecasting and vaccine production 

Although BioNeMo technology is promising, the accuracy and dependability of predictions and forecasts made by their AI systems may be impacted by incomplete or fragmented datasets, since its AI models rely significantly on high-quality training data. Furthermore, the intricacy of these models may restrict interpretability and might make it challenging for researchers to comprehend the logic underlying certain results.  

The technology needed to operate BioNeMo is another factor to take into account since it might be a deterrent for smaller businesses or university labs. However, as AI technology evolves and data security measures improve, these challenges could be mitigated, making BioNeMo accessible while ensuring greater protection against security risks and data breaches.