3d rendered illustration of the anatomy of a cancer cell: © Sebastian Kaulitzki - AdobeStock_305639707

The pharmacological universe stands at a critical junction right now. It's where the world of drug development intersects with artificial intelligence, and we're just starting to see the first fruits of research into this area. Going beyond wet lab trial and error, artificial intelligence engines designed for biologic precision are accelerating drug candidate identification, particularly in oncology.

Recognition of artificial intelligence in the area of pharmaceuticals received a significant bump this year when the Nobel Prize in Chemistry was awarded to scientists who developed an AI model capable of predicting the complex structures of proteins from their amino acid sequences.¹ In fact, that model has now been used to successfully predict the structure ofalmost all 200 million known proteins.²

Biologic-specific AI

While we're still in the early days of applying AI to drug design, the technology is already starting to transform the discovery of biologics. According to researchers based at the Boston Consulting Group (BCG), discovering new biologics is more challenging than designing other types of drugs because of the molecule size. Small-molecule drugs typically have sizes ranging from 200 to 700 Da, so AI can be used directly to design the optimized molecular structure. On the other hand, biologics typically have a size between 5,000 and 200,000 Da, which means current limitations in computing power and technology prevent direct, atom-by-atom design of biologics using AI.³

Despite those challenges, at least 50 to 60 AI-enabled biologics are already in various stages of discovery and development, a number expected to grow rapidly on the back of further advancements in AI technology, computing power, and data availability.³ Oncology is an area of particular interest in AI-enabled biologics discovery because current treatments destroy healthy cells alongside malignant ones. BCG identified at least eight AI-derived biologics in clinical trials for oncology.

As a result of the limitations in computing power, AI is used differently in biologics in five specific areas (as described by BCG): "target discovery and validation, target structure prediction and epitope definition, diversity generation and screening, humanization and engineering of biologics, [and] property prediction and optimization."

For example, AI is helping design better molecules, including antibodies, protein-based therapeutics and other biologics. Although direct, atom-by-atom design is not possible in biologics, the technology is enabling the identification of effective drugs in weeks instead of months. It also tests the developability of various drug candidates.

AI is also enabling researchers to screen women for early signs of ovarian cancer by detecting genetic changes and protein biomarkers in a blood test.⁴ Researchers at the Johns Hopkins Kimmel Cancer Center collaborated with several other institutions for a study using AI-powered analyses of DNA fragments and two protein biomarkers.

How a multiomics approach enables AI to create novel outputs

Of course, for AI models to work, they must first be trained on an immense amount of data. For example, the AlphaFold2 model developed by the recipients of the Nobel Prize in Chemistry was trained on all known amino acid sequences paired with their determined protein structures. However, that's just the first step.

The next logical step is taking that protein data and applying it to drug development, and multiomics provides a critical piece of the puzzle. After all, proteins are the building blocks of antibodies that protect us from disease-causing pathogens.

Multiomics is a technique used for biological analysis that leverages data from multiple datasets, like genomics, epigenomics, proteomics, microbiome, metabolome, transcriptomics, and more. Today, only the most sophisticated systems can efficiently and at scale synthesize sequence data, 3D structural biological models, formal knowledge bases and ontologies, and public text sources (like patient records).

However, this is an area in which we would expect significant growth in the coming years, as an increasing number of AI models begin to support a multiomics approach.

Multiomics solves a critical problem that is now facing the industry: an enormous amount of data spread across multiple sources. Using a multiomic approach, AI seamlessly integrates and harmonizes massive data from diverse sources to create new, thoughtful outputs that incorporate different aspects of biological systems. Multiomics also enable the development of novel compounds like bispecific antibodies that can address unique oncological targets.

Multiomics in action

Computer-aided molecular design enables large numbers of chemicals to be assessed at once to determine the binding between the drug candidates and targets while predicting their toxicity and pharmacokinetic profiles earlier in their development.⁵

Combining multiomics analysis with computational models to develop cancer treatments is currently a novel pharmacological paradigm in cancer. Computational approaches work with omics data to enable the identification of new therapeutic targets and individualized approaches in cancer therapy.

For example, cancers like non-small cell lung, breast, head and neck, colon, and pancreatic have demonstrated high tumor-associated antigen tropomyosin receptor kinase B (TrkB) levels, which are proteins associated with low survival rates and poor patient outcomes.

ImmunoPrecise Antibodies' (IPA)'sLENS^ai system generated multiple unique antibody candidates that target a protein with a previously unknown structure in the tumor microenvironment.⁶ The AI model also predicted and designed novel sequences with new binding patterns and activities. This led to the development of unique, proprietary, bispecific antibodies that link cells expressing the tumor-associated antigen TrkB with CD3-positive T cells, which are immune cells that respond by identifying and then eliminating cancer cells.

That research has enabled IPA to work on a potential therapeutic that's designed to bring TrkB-expressing tumor cells together with engaged, activated T cells, which could represent a novel, first-of-its-kind therapeutic for difficult-to-treat cancers.

Additionally, BostonGene has developed an AI-powered multiomics platform that optimizes drug development and clinical trial designs by identifying multiparametric signatures and enabling precise matching of patients with the therapies that will be most effective for them.⁷ As a result, the AI platform has sped up the R&D process, bringing personalized cancer care to an increasing number of patients around the globe.

BostonGene also played a major role in the TROP2-directed antibody-drug conjugate clinical trial with LegoChem Biosciences. Multiomics enables BostonGene to profile patients and enhance patient selection to improve and even maximize therapeutic outcomes.

The two biomarkers, cancer antigen 125 and human epididymis protein 4, were previously linked to ovarian cancer but could not reliably detect it on their own. The study found that applying AI to the problem and combining the biomarkers with AI-driven detection of patterns associated with cancer in DNA fragments improved the accuracy of screenings while helping tell cancerous tumors apart from benign growths.

Why AI won't replace wet-lab work in drug development

While AI advances are changing the world of pharmaceutical discovery and development forever, they will never eliminate the need for wet-lab work, which is nuanced and essential in helping biopharmaceutical companies develop effective, reliable, and relevant tools to enhance AI drug discovery.

Critical thinking, and especially human creativity, can never be fully replicated by AI due to the complexities of the natural world. In reality, combining AI with wet-lab work creates an endless cycle from the lab, to data, and then back to the lab, where testing occurs.⁸

A critical problem that will always exist for AI is a relative lack of pre-existing data on biology and the natural world. Scientists are constantly making new discoveries that require the types of critical thinking and creativity that AIs are not capable of, and researchers are always filling in more pieces of the puzzle. This generates new data for AI models to learn from, but those discoveries and the data on them must exist first before they can be used to train an AI.

Dr. Jennifer Bath is an immunologist and cellular and molecular biologist. She is also president and CEO of ImmunoPrecise Therapeutics.

References

1. "They cracked the code for proteins' amazing structures." The Nobel Prize. 2024. Available from:https://www.nobelprize.org/prizes/chemistry/2024/press-release. Press release.
2. The Royal Swedish Academy of Sciences. "They have revealed proteins' secrets through computing and artificial intelligence." The Nobel Prize. 2024. Available from:https://www.nobelprize.org/uploads/2024/10/popular-chemistryprize2024-4.pdf
3. Jayatunga M, Bruens L, Ruder L, et. al. "AI in biologics discovery: an emerging frontier." In Vivo. Accessed 26 Dec. 2024. Available from:https://web-assets.bcg.com/4b/f8/ef10f64643578c65e34902c81513/ai-in-biologics-discovery-an-emerging-frontier.pdf
4. "AI 'liquid biopsies' using cell-free DNA, protein biomarkers, could aid early detection of ovarian cancer.' John Hopkins Medicine. 2024. Available from:https://www.hopkinsmedicine.org/news/newsroom/news-releases/2024/09/ai-liquid-biopsies-using-cell-free-dna-protein-biomarkers-could-aid-early-detection-of-ovarian-cancer. Press release.
5. Fatima I, Rehman A, Ding Y, et. al. "Breakthroughs in AI and multi-omics for cancer drug discovery: A review." Eur J Med Chem. 2024 Dec 15; 280:116925. doi: 10.1016/j.ejmech.2024.116925. Epub 2024 Oct 4. Erratum in: Eur J Med Chem. 2024 Dec 5; 279:116968. doi: 10.1016/j.ejmech.2024.116968. PMID: 39378826.https://pubmed.ncbi.nlm.nih.gov/39378826/
6. IPA. IPA’s Subsidiary, BioStrand, Provides an Update on LENSai™, Oct. 25, 2023 Press release https://ir.ipatherapeutics.com/news/news-releases/news-details/2023/IPAs-Subsidiary-BioStrand-Provides-an-Update-on-LENSai-2023-xcA0JQr_0e/default.aspx
7. "Pharmaceutical Technology Excellence Awards 2024: BostonGene." Pharmaceutical Technology. 2024. Available from:https://www.pharmaceutical-technology.com/featured-company/2024-bostongene/. Press release.
8. Mallavarapu A. "The rise of 'wet' artificial intelligence." Proto.life. Nov. 2023. Available from:https://proto.life/2023/11/perspective-the-rise-of-wet-artificial-intelligence/