Issue 5: AI in Biologics | NVIDIA

Today we will cover a topic impacting the preservation of human life: Advancement of AI in the field of Biologics or Large Molecule drugs.

Let us start with a primer on the field of Biologics, for those of us that may be new to the space. If you are familiar, you may skip ahead to the other sections.

Background

Pharmaceuticals play a significant role in preservation of human life - from antibiotics to vaccines, research in drug discovery is focused on developing new ways to prevent and treat disease. At a very high (and simplified) level, Pharmaceuticals are organized into two categories: small-molecule drugs and large-molecule drugs.

·       Small molecule drugs are medications with a low molecular weight made up of chemicals created in a lab; they have a simple chemical structure; they are relatively easy to make and have high shelf life. They are generally administered as a pill or orally as a syrup, they can be easily absorbed into the bloodstream and may travel to all body parts, or cells, where they interact with other molecules within the cells. They make up a large segment of today’s OTC (over the counter) or prescription medications, providing solutions for common and chronic conditions. Most of us have regular interactions with small-molecule drugs like ibuprofen, antihistamines, or penicillin.

·       Large molecule drugs, have a large molecular weight and are manufactured or extracted from living organisms, such as a microorganism, plant cell, or animal cell. They are made of proteins that are complex in structure, costly to manufacture and store, and are mostly administered through IV or infusion. Biologics are the rising stars in medicinal drugs, offering treatments for complex diseases like cancer, autoimmune disorders, gene therapies, tissue transplants or stem cell therapies. To put it in perspective by numbers, while a small molecule drug, such as aspirin, can consist of 20-50 atoms, biologics can be made of 25,000-50,000 atoms. This category of drugs are also referred to as Biologics and the industry is referred to as Biopharma. It is worth noting that 7 of the top 10 selling drugs in the US are Biologics. Top Biologics manufacturers are in Figure 2.

Complexity and cost in Biologics:

Biologics are produced from living cells, such as bacteria, yeast, mammalian cells, or other cell types, that are genetically engineered to reproduce copies of the desired protein or antibody, to be used as a drug for treatment. The process is not only complex but requires time and high budgets, which are naturally transferred to the patient. Therefore any solutions to reduce the complexity or cost, would reduce the cost to the patients and improve access and affordability, which is the ultimate goal.

Here is a simplified view of the costs:

1.      Complex manufacturing process: Biologics are created from biological products which include sugars, proteins, nucleic acids, or complex combinations, which are more temperature-sensitive, require special storage, and have to follow FDA regulations throughout the development process.

2.      Clinical Trials: Patient recruitment and analyzing clinical trial outcomes and designing efficient trials requires substantial investments in time and money.

3.      Compliance: Biologics use living cells, which implies variability in each batch - manufacturers have to dedicate significant resources to ensure products behave predictably in all patients, meet regulatory compliance and mitigate risks throughout the process.

4.      Storage and administration: Cost of storing temperature-sensitive medications, providing space for infusions, and necessitating healthcare provider time, has added to the overall price of biologics in the U.S. healthcare system.

The lion’s share of Biologics drug discovery and manufacturing is focused on Cancer treatment, followed by Infectious Diseases (which rose to further prominence with Covid-19). Here is a chart in Figure 1:

AI, ML, Gen AI in Biologics

AI and ML have been used in Pharma, or Biologics for years, however the pace at which it is accelerating research and transforming the industry today, is unprecedented. For example, in one year, a human researcher can perhaps reviewing 300 articles while IBM Watson can process more than 25M article abstracts and 1M medical journal articles. From there, a variety of Applications can be developed. I have covered some below.

I have worked with customers in Pharma, as they all work with enormous data sets, that need to be cleaned, prepped and transformed for any downstream analytics or predictive modeling, but in this newsletter I cover Nvidia in particular. Nvidia is not a Biopharma manufacturer, but a critical enabler for the adoption of AI and Gen AI in the field of BioPharma.

With the advancements in LLMs and in Generative AI technologies, there is impact across the value chain. I found this chart to summarize the space well.

Gen AI use cases across the Bio Pharma value chain, here is a chart of explain in detail.

AI in biologics is dramatically transforming drug discovery, development and personalized medicine.

1.       Drug Discovery and Design: Today, researching, developing, and commercializing a medication takes an average of ten years. Clinical trials add another six to seven years. Bringing a single drug to market can cost Billions of dollars. With AI, they time may be reduced to 1-2 years (10%).

a.        Large datasets and a variety of data sources (such as genomics, proteomics, etc.) can be analyzed efficiently and fast, to predict the efficacy of specific compounds.

b.       Machine learning models can assist in designing new biologics. AI assists in the virtual screening of compounds to identify potential drug candidates, saving time and resources in the early stages of drug development.

c.        Today, many diseases simply have no treatment. With AI, we can generate therapies that can predictably treat patients very effectively.

2.       Operations - Generative AI brings increased speed and efficiency to every stage of the process, from discovery to adoption.

a.        The ability to process (and learn from) terabytes of structured and unstructured data (such as academic literature, clinical-trial information, batch records, quality investigation reports, electronic medical records, and claims databases)

b.       The ability to generate completely new content: text, media (image, speech, and video), design, and code

c.        Intuitive, conversational user interfaces that reduce AI’s adoption hurdles, making the technology widely accessible.

3.       Clinical trials: AI modeling can identify suitable patient populations for the trials, and improve efficiency and accuracy of patient responses.

4.       Personalized medicine: AI Modeling can analyze genetic profiles to tailor treatment plans for patients, while minimizing side effects. In the near future, we will have pre-trained data from LLM’s and individual patient data from different inputs like wearables or EMR systems. Now the models can predictably provide personalized treatment, from figuring out not just what drug might work for a patient but exactly what drug would work when, in what sequence, in what dose.

However, it's important to note that regulatory and ethical considerations play a crucial role in the implementation of AI in the biopharmaceutical industry. I will be covering this in an entirely new newsletter.

Enter NVIDIA

Reference: BioNeMo | Generative AI Platform | NVIDIA

NVIDIA currently dominates the market for chips used for AI applications. Earlier in 2023, NVIDIA announced BioNeMo, to accelerate the adoption of generative AI, and accelerating the most time-consuming and costly stages of drug discovery. Generative AI models are trained on large-scale datasets of small molecules, proteins, DNA and RNA sequences, spanning a variety of data types (language, images, video and 3D) that can then predict the 3D structure of a protein and how well a molecule will dock with a target protein.

These pre-trained models can they be applied to proprietary data, helping researchers identify the right target, design molecules and proteins, and predict their interactions in the body to develop the best drug candidate. It is dramatically accelerating years of manual research and drug discovery.

BioNeMo has six open-source models:

• AlphaFold2: A deep learning model that reduces the time it takes to determine a protein’s structure from years to minutes or even seconds, just by using its amino acid sequence, developed by DeepMind and already used by over a million researchers.
• DiffDock: To help researchers understand how a drug molecule will bind with a target protein, this model predicts the 3D orientation and docking interaction of small molecules with high accuracy and computational efficiency.
• ESMFold: This protein structure prediction model, using Meta AI’s ESM2 protein language model, can estimate the 3D structure of a protein based on a single amino acid sequence, without requiring examples of several similar sequences.

• ESM2: This protein language model is used for inferring machine representations of proteins which are useful for downstream tasks such as protein structure prediction, property prediction and molecular docking.
• MoFlow: Used for molecular optimization and small molecule generation, this generative chemistry model creates molecules from scratch, coming up with diverse chemical structures for potential therapeutics.

• ProtGPT-2: This language model generates novel protein sequences to help researchers design proteins with unique structures, properties and functions.

Nvidia BioNeMo collaborates with AWS and Oracle in addition to Google DeepMind and Meta AI as noted above.

Use cases and customers

·        Amgen was able to slash the time it takes to train five custom models for molecule screening and optimization from three months to a few weeks. It trained BioNeMo’s models using its own proprietary data on antibodies

·        Evozyne’s proprietary protein data was trained by BioNeMo’s training sets, and used to design synthetic variants with significantly improved performance compared to enzymes found in nature.

·        Insilico Medicine has reached a new milestone in drug discovery designed by generative AI to treat Idiopathic pulmonary fibrosis (IPF), a rare lung disease with irreversible decline in lung function

·        Exscientia enables scientists to use Gen AI models to design a new drug for Alzheimer’s disease, that has few side effects. A well-trained ML Model could generate molecules that have never been synthesized before, but have the potential to be more effective and lss toxic than existing drugs.

·        Adaptyv Bio, is creating novel protein designs whether that's an antibody, an enzyme or something completely new.

·        Absci ambition is to grow into the Google search engine of protein-based drug discovery and biomanufacturing. Absci says its AI-designed antibodies can cut drug discovery timeframes by more than.

NVIDIA is a critical player in the adoption of AI in BioPharma. They are also profilic in supporting the startup ecosystem to accelerate innovation and growth, having invested in 35 AI Companies just in 2023. To better understand the entire startup landscape, you can look at Figure 3 below.

Charts

Figure 1:

Figure 2.

Figure 3:

Additional reading:

AI in biopharma research: A time to focus and scale | McKinsey

AI in Biologics Discovery: An Emerging Frontier (bcggoo.com)

11 drug discovery startups to watch, according to VCs  | Sifted

New Funding Rounds in Biologics | VentureRadar