Biotech Wasn't Ready For AI's Speed; Here's How We Catch Up

By Lee C. Smith and Andrew M. Thomson, GreyRigge Associates Ltd.

The marriage of artificial intelligence (AI) and biotechnology is now moving from theory to practice, which is becoming abundantly clear, given where the venture capital investment is flowing globally within biotech.¹ Notably, much of the funding is going to companies like Isomorphic Labs, Insilico Medicine, Recursion Pharmaceuticals, and Generate Biomedicines, as well as huge investments in AI in general.^2,3

In biotech, the AI input ranges from de novo molecule design to systems biology, with AI tools increasingly being embedded in discovery and development workflows. While algorithms have been used in predicting a molecule’s folding for decades, integration of AI into standard workflows is a disruptor, altering how targets are identified, how therapeutic candidates are prioritized, and how risks are assessed throughout the product development life cycle.

In simple terms, the biggest bottlenecks in creating a commercial product are the development of manufacturing processes together with demonstrating that the product is nontoxic, safe, and efficacious as demonstrated within clinical trials. Therefore, innovators are seeking to utilize AI to remove these bottlenecks.

However, the pace of capital deployment has, in many cases, outstripped the sector’s readiness to critically evaluate and de-risk these programs, which is a major error. While computational methods may currently yield promising predictions based on some previous training data, the biological relevance along with the ability to manufacture and the regulatory feasibility of these outputs remain critical constraints and weak elements in otherwise compelling investment cases.⁴  

If we go back 20 years, one of the seminal papers in microRNA predictive expression and targets demonstrated clearly the need for validation in the biological context.⁵ This article examines how traditional technical expertise in the product development quality by design (QbD) portfolio, such as assays, CMC, preclinical strategy, and regulatory planning, must be harnessed and evolved to support and validate AI-driven development. The ongoing revolution in human organoid models will accelerate AI validation, as well as regulatory acceptance, and will be an important driver.

Validation of AI tools also has implications for biotech companies and regulatory bodies and their capability to predict efficacy and toxicology as well as long-term safety of the product they develop. This includes massively complex drugs such as gene and cell therapies, cell-derived extracellular vesicles, and secretomes. Either way, we consider that the AI used in biotech must be driven by data validating these predictions.

Emerging AI-Biotech

The landscape of biotechnology is rapidly being reshaped by organizations that incorporate machine learning (ML) and AI into the biopharmaceutical development foundations. These focus on:

• molecular properties prediction for optimizing leads in silico
• target discovery using omics, imaging, and patient-derived data
• de novo design of therapeutic constructs, for example, peptides, mRNA, ncRNAs, or small molecules
• digital biomarker discovery and patient stratification tools
• cell culture optimization
• product formulation
• PK predictions
• DOE optimization.

The goal remains to compress the timelines of discovery and development as well as improve the efficiency of development, whether by reducing costs, reducing timelines, or increasing productivity. This is achieved by improving the likelihood, based upon AI predictions, of generating novel therapeutics with higher probabilities of clinical and commercial success.⁶ High-profile partnerships, early-stage licensing deals, and public listings have followed. Yet many of these ventures remain exposed to critical risks. This isn’t due to their algorithmic logic but to the risk of them potentially having poor biological translation.⁷

AI Output Does Not Mean Biologically Ready

It is possible that problems with AI-powered biotech solutions may arise from assuming that computational success realistically aligns with therapeutic readiness. Models trained on well-annotated data sets may yield novel outputs such as target predictions, molecular structures, or biomarker signatures. However, such outputs must still pass through the familiar and unforgiving crucible of experimental validation, manufacturability assessment, and regulatory engagement.⁸

These challenges raise a host of critical questions:

1. What is the origin and quality of the training data?
   Is it sufficiently diverse, clinically relevant, and unbiased?
2. Are AI-derived outputs biologically plausible and do they provide real-world insights?
   Do the predictions align with known mechanisms, pathways, or disease models? If not, does this lead to new discoveries and further validation?
3. Can outputs be translated into real constructs and be manufactured?
   Can novel proteins, antibodies, or RNA constructs be made reliably and cost-effectively?
4. What are the regulatory implications of using AI in this context?
   Does the approach trigger additional regulatory scrutiny, e.g., a requirement to screen off-target effects, such as, for example, CRISPR induced changes and mRNA dysregulation that drive cancers, which may require further algorithm development based on clinical readouts?
5. How should AI-generated insight and validation data be shared?
   If validation data is shared, the AI developers benefit, but biotech companies and investors will question that they’re subsidizing model development for the benefit of others?  If the biotech’s generate data to confirm AI outputs, who then owns the IP?  Is it shared? Why should biotech companies pay AI companies to validate unvalidated AI predictions?

These questions should challenge both innovators and investors seeking to differentiate promising platforms and technologies that are rooted in sound science from those technologies or platforms that are reliant on speculative extrapolation.⁹

Expanding The Due Diligence Framework

Conventional diligence practices will evolve to accommodate the unique attributes of AI-driven platforms. Historically, due diligence has commonly focused on target validation, intellectual property, CMC feasibility, preclinical data, and regulatory planning. In the AI context, these must now be supplemented by the following:

1. Data lineage and integrity
   Understanding the origin, review, and how representative a data set is essential. Models trained on narrow or unbalanced data sets may result in biased predictions with limited real-world utility.¹⁰ In addition, attention must be paid to data licensing, privacy, and consent. This is critical if any patient clinical data is involved.
2. Biological plausibility and reproducibility
   AI predictions must be reconciled with established and constantly updated complex biological and clinical frameworks, especially with deconvolution capability of massively complex data sets. The presence or absence of independent experimental corroboration with in vitro, in vivo, ex vivo, and clinical data is essential as these all form the foundations of validation.¹¹
3. CMC feasibility
   If AI is used to optimize molecules or design entirely novel constructs, then the question of CMC compatibility is raised. Studies around construct stability, expression, or product toxicity to cells will still need to be performed. It may also impact the ability to formulate the product or its stability. Therefore, this needs to be considered for all assay and manufacturing data to improve rapid design and operation space and not only in the drug’s design. All of these aspects are required to aid AI design to proceed through the manufacturing pipeline without prohibitive technical risks.^12,13
4. Regulatory positioning
   Where AI plays a role in patient selection, biomarker identification, or real-time decision support, classification as a software as a medical device (SaMD) or a combination product may apply .¹⁴ Indeed, does clinical data analysis (e.g., miRNA prognostics and diagnostics) also aid AI design for more efficient clinical trial design? Understanding how regulators interpret such innovations will become increasingly important to preempt misalignment or misunderstandings that lead to delays.

Implications For Investors

The increased complexity introduced by AI platforms requires investors to interrogate not only the potential of the technology but also the discipline with which it is being applied. Important questions will be:

• Has the company defined a clear experimental plan to validate its computational outputs?
• Are there domain experts embedded in the team or an advisory structure with sufficient biological, manufacturing, regulatory, and clinical experience?
• Is the AI component integral to the therapeutic value proposition or merely an enabling tool?
• What are the barriers to laboratory validation, and how are they being addressed?
• Are there technical development milestones that will help with validation and ultimately de-risking of the platform?
• If lacking internal expertise, advice from external subject matter experts can be used to provide a critique to ensure that the enthusiasm for AI is counterbalanced by a rigorous examination of its feasibility.¹⁵

Implications For Innovators

While it’s still early days, AI biotech companies to date appear to have mainly emerged from academic, discovery, or engineering-led backgrounds. While this brings expertise and innovation in data science, it may leave critical gaps in development planning. It is important that innovators recognize that:

• The evidence required for partnering or fundraising includes robust biological validation.
• Investors will expect de-risking road maps to exist to demonstrate technical maturity, not just model optimization.
• A clear manufacturing and regulatory strategy must accompany the scientific vision, especially for novel modalities.

It is important to start early in defining validation experiments, structure CMC plans and anticipate the regulatory questions, all of which contribute direction to valuation and partner interest.¹⁶

Cross-Disciplinary Collaboration Matters

Supporting the successful translation of AI biotech innovations requires an integrated team, including:

• data scientists to explain model architecture, bias, and interpretability
• biologists and pharmaceutical developers to assess mechanistic credibility and hypotheses that can be validated
• CMC experts to evaluate ability to manufacture, develop assays, and formulate
• regulatory professionals to guide interactions with agencies
• clinical developers to plan early-phase studies with appropriate endpoints.

This multi-dimensional evaluation is a rare combination seen within a single company, especially at early-stage development. External subject matter expertise (SME) applied at the right time may be the difference between a promising concept and a development-ready program.¹⁹

Delivering On The Promise Of AI-Powered Biotech Solutions

To conclude, AI-powered biotech solutions are here to stay, offering the potential to both accelerate and refine the discovery of new therapeutics, improve efficiency in CMC and clinical trial design, as well as aid regulatory guideline development. However, the path from algorithm to licensed product will likely remain slower than desired, as well as remaining complex. This is because of the confirmatory exercises required. Success depends not only on the power of the models but on the ability to validate, manufacture, and demonstrate safety and efficacy while bringing regulatory agencies on board with what modelling produces.

For investors, the ability to interrogate AI-derived opportunities through the lens of biology, CMC, and regulatory science is hugely attractive. For innovators, early engagement with technical product development experts ensures that platforms are built not only to predict but to also translate. This should reduce onerous outcomes that impede regulatory agency capabilities to move efficiently and with confidence while also reducing the timeline and cost of bringing products to market.

As the fusion between AI, CMC, and clinical development continues to mature, a disciplined, cross-functional approach, including QbD, to evaluate opportunities for AI-powered biotech solutions will grow in importance to ensure that future novel products are computationally elegant, efficient, and cost-effective to manufacture and clinically meaningful to patients.

References:

1. Olsen, E. “Healthcare venture capital investment boosted by AI in 2024: report.” Biopharma Dive, 2025. https://www.biopharmadive.com/news/healthcare-venture-captial-funding-ai-boost-2024-silicon-valley-bank/736902/
2. Teare, G. “Startup Funding Regained Its Footing In 2024 As AI Became The Star Of The Show.” Crunchbase News, 2025. https://news.crunchbase.com/venture/global-funding-data-analysis-ai-eoy-2024/
3. BIO Industry Analysis. “Artificial Intelligence in the Biopharmaceutical Industry.” BIO, 2023. https://www.bio.org
4. Zeggini, E., et al. “Translational challenges in AI-driven biology.” Nature Reviews Genetics, 2023. https://www.nature.com/articles/s41576-023-00577-5
5. Miranda, K.C et al. “A Pattern-Based Method for the Identification of MicroRNA Binding Sites and Their Corresponding Heteroduplexes Cell.” 2006 126(6):1203-17 S0092-8674(06)01099-3
6. Beam, A.L. & Kohane, I.S. “Big Data and Machine Learning in Health Care.” JAMA, 2018. https://jamanetwork.com/journals/jama/fullarticle/2675024
7. Nature Biotechnology Editorial. “AI in Biotech: Too Much Too Soon?” Nat Biotechnol 2023. https://www.nature.com/articles/s41587-023-01958-2
8. FDA. “Artificial Intelligence and Machine Learning (AI/ML) in Software as a Medical Device.” Draft Guidance. https://www.fda.gov/media/155022/download
9. European Medicines Agency. “Reflection paper on the use of AI in the medicinal product lifecycle.” EMA/257600/2023
10. Markus, A.F., et al. “Data bias in AI: a threat to model validity.” PLOS Medicine, 2021. https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1003653
11. Ekins, S., et al. “Exploiting machine learning for end-to-end drug discovery.” Drug Discovery Today, 2019.
12. ICH Q8-Q12 Guidelines. International Council for Harmonisation. https://www.ich.org
13. FDA. “Assay Development for Cell and Gene Therapy Products.” 2022 Guidance.
14. MHRA. “Software and AI as a Medical Device: Change Programme Roadmap.” 2022.
15. Deloitte. “AI in Biopharma: Investing Wisely in a High-Stakes Environment.” 2023.
16. Kidney Precision Medicine Project. “Data Resources and Collaborative Opportunities.” https://kpmp.org
17. Himmelfarb, J., et al. “The current and future landscape of kidney disease research.” JASN, 2021.
18. PricewaterhouseCoopers. “Accelerating AI in Health: The Need for Multidisciplinary Expertise.” 2023. https://www.pwc.com

About The Authors:

Andrew is a seasoned biotech consultant with more than 30 years of global experience in cellular therapies, vaccines, RNA and antibody therapeutics, and biodefense. A UK-trained molecular biologist and co-founder of GreyRigge Associates, he leads CMC and process development strategies that translate complex science into accelerated product and regulatory success for clients worldwide.

Lee Smith founded GreyRigge Associates in 2010 after holding senior roles, including CEO- and VP-level positions. Lee has worked for small biotech startups and multinational corporations in the U.S., U.K., and Singapore for companies such as GSK, Emergent BioSolutions, and SingVax. 

Contact them at contactus@greyrigge.com. Learn more about GreyRigge at www.greyrigge.com.