Leveraging GPU and AI for Biotech Innovation

GPUs have revolutionized AI and computing through three primary strengths:

• Parallel processing capabilities.

• The ability to scale up to supercomputing levels.

• Enabling robust software ecosystems designed specifically for AI applications.

These features enable GPUs to handle complex calculations with greater speed and energy efficiency than CPUs, making them the preferred choice for AI training, inference, modeling, and other high-performance computing tasks. Unlike CPUs, which process data sequentially, GPUs use parallel processing to handle large datasets and complex mathematical models more efficiently. This parallel processing is especially important in biotech, where GPU and AI technologies are enabling organizations to analyze protein folding or genomic interactions that require handling enormous amounts of data in real-time.

How much of a difference are GPUs making in AI? Stanford’s Human-Centered AI group found that GPU performance gains have nearly tripled the computational speed since 2021. This surge in GPU performance is helping researchers develop more accurate, scalable models that process data at unprecedented speeds, accelerating the rate of discovery across biotech and genomics.

In 2024, NVIDIA’s BioNeMo and DGX Cloud platforms incorporated innovative GPU capabilities that enabled faster, more scalable AI model training. Furthermore, with the recent release of NVIDIA’s Grace Hopper Superchips, which are touted to deliver 10X higher performance for applications running terabytes of data, GPU and AI technologies are further accelerating biotechnology model processing and making them more energy-efficient. These new GPUs are also optimized for the complex matrix math that neural networks use, so they are ideal for biotech applications like protein modeling and genomic sequencing which rely on high-performance, large-scale calculations.

Now that we’ve covered the basics of how GPUs are accelerating AI, let’s look at the practical applications of GPU and AI for biotechnology organizations.

LLMs for Genomic Research

The 2022 ACM Gordon Bell Special Prize recognized a team of researchers for groundbreaking work in modeling virus evolution. The team created powerful large language model (LLM) AIs, known as genome-scale language models (GenSLMs), which were trained on over 110 million gene sequences and fine-tuned with millions of SARS-CoV-2 genomes. These models were a major breakthrough and demonstrated impressive scaling capabilities on supercomputers like Polaris and Selene where they were used to simulate and analyze viral mutations and immune escape mechanisms.

Leveraging LLMs in genomics has unlocked new ways to understand the context and interactions of genetic sequences across the genome. This is a great example of how AI is being scaled up and enhanced and will continue to break new ground using advanced models like GPT-4 that are trained on supercomputers.

The Expanding AI Applications in Biotech

One of our favorite biotech generative AI platforms is the NVIDIA BioNeMo. This platform is powering advanced drug discovery and genomic research, and it now includes microservices tailored to various stages of drug development, such as protein interaction analysis and molecular simulation, which can accelerate therapeutic research. Operating on NVIDIA's DGX Cloud, BioNeMo’s microservices enable scalability and are instrumental in streamlining the analysis of protein structures and interactions.

Biotech organizations are also leveraging AI and GPUs to create robust models that analyze and predict protein folding and molecular interactions that are critical to understanding disease mechanisms and accelerating drug design. Amgen was able to use the platform for rapid training and fine-tuning of LLMs on proprietary protein data. It resulted in generative models that can predict protein properties and generate candidate molecules that quickly meet therapeutic criteria and reduce lab testing requirements. This has enabled Amgen to achieve protein structure predictions in as fast as 20 seconds and complete training for initial models in under four weeks, significantly speeding up R&D!

In addition, Google DeepMind’s AlphaFold3 is predicting protein-DNA and protein-RNA interactions and 3D structures of molecular complexes including large biomolecules like proteins, DNA, RNA, and small drug molecules. Google claims, “For the interactions of proteins with other molecule types we see at least a 50% improvement compared with existing prediction methods, and for some important categories of interaction we have doubled prediction accuracy.” The AlphaFold Server, freely available for non-commercial research, enables scientists to rapidly predict complex molecular structures that could save years of experimental work and hundreds of thousands of dollars.

How Can Biotech Organizations Leverage GPU and AI?

As you can see in the examples above, these GPU and AI technologies are helping biotech organizations accelerate results in many areas. Some of the low-hanging fruit use cases you should be considering are:

• Training AI models: Biotech organizations can harness the immense parallel processing capability of GPUs to train complex AI models on vast datasets, dramatically reducing the time and cost associated with computational tasks like protein structure prediction, genomic analysis, and molecular simulation.

• Drug discovery and predicting protein structure: AI solutions can accelerate drug discovery by enabling detailed visualization of molecules and their interactions with proteins. This allows scientists to observe the actual molecular structures, see how they attach, and test their stability under simulated conditions—providing crucial data for selecting the best protein candidates for trials. Additionally, one of our clients leverages BioNeMo’s GPU and AI-powered technology to predict how proteins will change shape in response to a drug molecule, and it has been transformative for their scientists.

• Biomolecular workflows: By using GPU-driven computational power, biotech organizations can quickly analyze and model complex biological data, making it possible to predict molecular structures, simulate protein-ligand interactions, and explore the stability and efficacy of potential drug candidates. You can also design workflows that accelerate critical stages in the drug discovery process, from initial molecule selection to testing protein interactions under various conditions. This enables scientists to work with large biological datasets in a streamlined, high-performance environment. 

• Analyzing DNA sequences: GPU and AI technologies facilitate rapid processing of DNA sequences and process data from cryo-electron microscopy (cryo-EM). This enables researchers to determine the intricate 3D structures of biological molecules, such as proteins. Scientists can leverage these technologies to capture high-resolution images of molecules at ultra-low temperatures to preserve the natural state without the need for crystallization. Researchers can use AI to perform the third step in structure determination, where the 2D cryo-EM images are assembled into a 3D model. This high-precision modeling is critical for understanding molecular functions, interactions, and stability, providing valuable insights that can guide the development of new drugs and therapies. 

Now that you’re thinking about use cases, let’s look at how you can implement these technologies within your organization.

Implementing AI in Your Organization

As you can imagine, implementing these technologies and platforms requires a lot of strategic planning to ensure your data, storage, and security solutions are designed to integrate seamlessly. How you select and implement these technologies makes a dramatic difference in the efficacy and results you’ll achieve. We recommend that you:

1. Evaluate AI platforms and tools that will deliver reliable, impactful results. You’ll want to consider the different features, security, and integration capabilities.

2. Design a data-driven storage solution. Advanced biotech research generates massive amounts of data. This will likely need to be a hybrid cloud strategy to maximize your speed while staying within your budget. You’ll need to determine how much data (and what level of sensitive IP data) should be stored on-premises for steady-state processing versus more affordable cloud storage. Look for solutions that are secure and scalable, then carefully (and regularly) review the configuration settings to secure your sensitive data and avoid a breach.

3. Acquire the appropriate network and security technologies. The best fit for your organization will depend on your needs, environment, size, and stage. Ensure you tailor the environment for the unique demands of biotech AI, ensuring compatibility, performance, and reliability.

4. Ensure seamless platform integration. We recommend paying close attention to your balance of cloud and on-premises platforms for smooth, efficient workflows, allowing your systems to communicate seamlessly and accelerating data processing.

If implementing and integrating AI into your organization feels overwhelming, we can help. At Pennant, we specialize in biotech IT solutions and we offer everything from AI consulting and implementation to cybersecurity and scalable storage and infrastructure solutions. We ensure your environment can handle the massive data demands of advanced biotech research and ensure fast, secure solutions. Our expertise in integrating cloud and on-premises platforms enables seamless workflows that enhance data processing efficiency, and we help you implement solutions that optimize security while maximizing performance within your budget.

We understand the unique challenges of the biotech industry and deliver solutions tailored to your specific goals. Please contact us if you need help implementing IT or AI solutions that are robust, secure, efficient, and deliver tangible results.