NVIDIA Leverages AI to Transform Drug Discovery and Development

Drug development is a critical process in the discovery of new medicines. The task of researchers during drug development is to identify novel molecules with positive therapeutic effects that can help address current health challenges.

However, it is no exaggeration to state that addressing the complex interplay among potential genes, biochemical reaction pathways, the chemical properties of protein targets, and absorption, distribution, metabolism, and excretion (ADME) processes is highly challenging. In summary, successful drugs must be capable of modulating targets in a highly specific manner within disease pathways. In drug development, our goal is to identify novel small molecules that act on drug targets to exert inhibitory effects within specific pathways, thereby improving patient therapeutic outcomes while simultaneously reducing adverse side effects.

Leveraging NVIDIA GPUs, pharmaceutical and biotechnology companies can conduct AI-driven drug discovery and development, thereby streamlining their R&D processes. With these technologies, researchers can transform vast patient datasets into clear, actionable insights, identify personalized and precise druggable opportunities, and predict potential responses to new drugs.

Understanding the biological mechanisms of diseases is crucial for developing effective treatment regimens. Disease pathways constitute a component of these biological mechanisms, potentially involving tens of thousands of proteins that interact and provide feedback through complex mechanisms. Ultimately, we must identify druggable

Protein Targets of Approved Drugs. By analyzing the human genome, we have gained valuable insights into pathogenesis and facilitated the identification of novel protein targets along disease pathways.

By determining protein structures, we can derive the three-dimensional structural distribution of amino acid sequences. These 3D structures provide insights into the physical properties of small molecules suitable for drug development, such as their shape and characteristics. Protein structures are typically determined using forms of electron microscopy, such as cryo-electron microscopy.

(cryo-EM). Over time, leveraging physics-based computational methods and AI approaches, it has become possible to accurately predict the 3D structure of proteins using only limited data derived from amino acid sequences and nuclear magnetic resonance (NMR).

The field of chemistry is vast and all-encompassing; thus, the process of searching, organizing, and creating chemical databases is referred to as cheminformatics. With protein targets and their 3D structures in hand, researchers can proceed to conduct computer-aided virtual screening to identify molecules with appropriate chemical properties that favor binding to the target. Researchers search the chemical space by calculating physical property vectors for each molecule, known as “fingerprints,” which can be viewed as coordinates for individual molecules within the chemical space. Leveraging fingerprints and related embedding methods, practitioners can widely apply techniques derived from data mining and machine learning, such as clustering algorithms, molecular property prediction, and the development of quantitative structure-activity relationship (QSAR) models.

Molecular simulations can provide key physical insights into the potency of drug candidate molecules based on their interactions with protein targets. We can perform simulations at varying levels of precision depending on the required throughput. For example, through docking, we can gain information about physical interactions

a rough description, and since the computational work has been simplified, it can support the screening of billions of compounds. With the aid of free energy perturbation (FEP) methods, molecular dynamics, and quantum methods, the picture of interactions is becoming increasingly accurate, allowing synthetic chemists to derive essential information on feasibility/infeasibility.

An increasing number of companies and researchers are leveraging AI to enhance current drug discovery and development methodologies. Molecular simulations, such as docking, free energy perturbation (FEP), and molecular dynamics, require substantial computational power. Researchers are integrating AI approaches across various stages of drug development to accelerate the process. Predicting protein 3D folding states, ligand-target binding energies, and pharmacokinetic properties such as toxicity and absorption represent only the tip of the iceberg in current AI method development. Novel AI methods can advance research progress and yield results comparable to those obtained through costly computational methods, thereby optimizing existing workflows. By running simulations with AI, researchers can reduce costs and discover novel potential drug candidates at a faster pace than ever before.

NVIDIA enables GPU-accelerated computing and AI software to advance computational workflows in drug discovery, thereby helping to accelerate research. By leveraging NVIDIA-powered platforms and software, researchers and pharmaceutical companies can develop and deploy intelligent applications to enhance analytical and computational capabilities throughout the research process. With NVIDIA Clara Parabricks, researchers can accelerate genomics workflows. Using NVIDIA RAPIDS, researchers can perform GPU-accelerated DataFrame operations within Python Notebooks in a unified environment, seamlessly integrating with machine learning algorithms such as clustering, dimensionality reduction, and list tasks. In practice, GPUs can accelerate nearly all critical computational chemistry codes, including docking and free energy perturbation (FEP), as well as molecular dynamics and quantum electronic structure calculations.

To learn more about NVIDIA’s healthcare and life sciences solutions, please click herePro TopicPage.