Healthcare and Life Sciences
Advancing Speed, Accuracy, and Scale in Drug Discovery
Objective
Using NVIDIA GPUs, RAPIDS, NVIDIA BioNeMo™, and Amazon Web Services (AWS), Innophore is revolutionizing how we understand protein-ligand interactions at a molecular level, driving advances in the speed and accuracy of drug development.
Customer
Innophore
Partner
AWS
Use Case
Accelerated Computing Tools & Techniques
Products
NVIDIA BioNeMo
NVIDIA DGX
Learning the Language of Molecules
The traditional drug discovery process is lengthy, complex, and expensive, taking up to 15 years from discovery to approval. And the failure rate still remains high—90% of drugs in development fail largely due to limited efficacy, toxicity, or undesirable side effects.
A key difficulty in predicting side effects before entering pre-clinical studies is that drugs bind to similar sites across very different protein structures. This can lead to off-target effects, where a drug binds to molecules or proteins it wasn’t intended for.
Identifying such functional correlations between biomolecules like proteins and nucleic acids is a complex task. These specific molecules carry out a myriad of functions essential to life, and are the key to unlocking the discovery of successful novel drugs.
Austrian-based NVIDIA Inception member Innophore is unlocking new speeds, accuracy, and scales in drug discovery and biotechnology with its Cavitomix™ platform. The platform leverages the power of catalophores, ultra-large deep neural networks, and diffusion models to encode the “language” of proteins and nucleic acids into a digital format.
Using its catalophore technology to identify hotspots on biomolecules where key interactions occur, Innophore compares the physio-chemical properties of binding sites, predicting the off-target effects of drugs with greater accuracy. This reduces the risk of adverse side effects, improves drug safety, and increases the success rate of developing new therapeutics.
A large-scale study conducted by Innophore found that of 467 drug targets screened for off-target effects, its model can correctly identify on average 30% of the alternative targets determined by experiments within the top ten ranked hits. The ligand-agnostic model addresses the limitations of existing tools by detecting unexpected off-targets, meaningfully reducing risk in the drug discovery pipeline and improving candidate selection based on off-targets identified earlier in the process.
Combining BioNeMo with its catalophore technology, Innophore was able to precisely represent the binding sites in the entire human organism’s protein structures within two weeks—a task which would previously have taken more than a year.
Understanding the digital language of biomolecules unlocks the ability to predict and analyze proteins and nucleic acids, mapping the vast universe of biomolecular interactions and activities. This guides researchers toward more targeted and efficient applications, and can open up new avenues for designing and optimizing biomolecules for specific needs. Using Innophore’s AI supermatcher technology, the most promising molecular candidates are prioritized, driving higher accuracy and reducing time to insight.
Summary
- Drug discovery can take a very long time, and even then, it is more than likely to fail due to unexpected side effects.
- Innophore’s Cavitomix platform decodes the language of proteins and nucleic acids into a digital format, allowing researchers to understand molecules—and how they interact—at an incredibly granular scale.
- Combined with the compute power of NVIDIA GPU and BioNeMo framework, Innophore can run millions of molecular computations within seconds to predict how proteins will react, which is 100 to 1,000 times faster than previously possible.
- With this information, researchers can accelerate drug discovery at an unprecedented speed, accuracy and scale, while using much less energy and cost to do so.
Unlocking New Levels of Speed and Precision
Integrating power from NVIDIA’s T4 and V100 GPUs, Innophore developed a regression model built on convolutional neural networks (CNN). Innophore trained the CNN with thousands of catalophores representing protein-protein binding sites, crucial for identifying and predicting how molecules interact with each other and enabling precise molecular design and analysis.
On-demand cloud capacity from AWS has empowered Innophore to make leaps in speed. Employing AWS cloud systems, Innophore’s AI algorithms reduced computation time considerably, from 2,000 CPU hours down to 15 minutes per protein-protein binding affinity prediction, and just 5% of the original cost. With its vast array of instances deployed, Innophore has spent 270,000 NVIDIA GPU hours on AWS. Saving time and budget on in-house infrastructure, and reinvesting resources into boosting speed, scale, and accuracy, Innophore can now make millions of computations per day, 8,000 times faster than by using CPUs only.
This enabled Innophore to evaluate more than 15 million new variants of SARS-CoV-2. With the many resource savings made by leveraging AWS and NVIDIA solutions, Innophore can predict the emergence of new virus variants well before they get dominant, like the Omicron variant of SARS-CoV-2, giving pharmaceutical companies and policymakers a critical head start in their response efforts.
Solution
NVIDIA GPUs on AWS
- Large-scale molecular dynamics simulations of protein-protein interactions
- Innophore used 270,000 NVIDIA GPU hours to generate training data for a neural network that predicts binding affinities for millions of protein variants
NVIDIA BioNeMo
- Flexible utilization via API
- Allowed rapid generation of atomistic structures for more than 42,000 human proteins using AlphaFold, OpenFold, and ESMfold
- Innophore built this comprehensive structural collection of the human proteome as the basis for predicting binding sites and generating highly descriptive catalophores used for virtual off-target screening
Results
- A far greater understanding of proteins and their interactions is crucial for the development of novel drugs
- From 2,000 hours to 15 minutes for predictions, at just 5% of the original cost to run predictions
Molecular Transformation Across The Industries
Innophore’s developments in biomolecular research aren't restricted to healthcare applications.
In addition to the advances Innophore has made in drug discovery and safety assessments, it has also used its technology to identify and optimize enzymes for various industrial applications, like the development of more efficient, eco-friendly laundry detergents that perform better while using less energy and water.
Its understanding of biomolecular function is driving innovation in the development of complex molecules, including those used in high-value fragrances, resulting in higher yields, lower costs, and a reduced environmental impact.
Innophore’s ability to decipher the molecules of life is delivering tangible benefits across multiple industries, promising a healthier, more sustainable future.
“NVIDIA boosted our performance so that we can run five million off-target predictions per second. We’ve been at a few hundred before.”
Christian C. Gruber
CEO, Innophore
Ready to Get Started?
To learn more about NVIDIA solutions for healthcare and life sciences, contact us.