AI-Developed Drugs: Demystified

Decoded Drugs #2

What is an AI-Developed Drug? Let’s Define it Properly

AI-driven drug discovery has been growing at an unprecedented pace. As of 2022, the combined pipelines of 20 AI-focused biotech companies contained approximately 160 disclosed discovery programs and preclinical assets1. By 2023, 75 molecules had entered clinical trials, with 67 still in progress2. But does this mean AI is truly 'developing' drugs? While AI plays an increasing role in various stages of drug discovery, from target identification to clinical trial optimization, the term 'AI-developed' remains ambiguous. One drug, BXCL501, has reached the market through AI-assisted repurposing, but does that truly qualify as an AI-discovered and developed drug?

Here’s my take: A drug developed using AI/ML is one where an integrated AI/ML platform has played a significant role throughout the drug R&D process, reducing both the cost and time of development compared to traditional methods. This approach extends beyond using AI in discrete steps—it involves holistic integration across multiple areas of the drug development lifecycle, including:

• Target identification and validation
  ‍
• Lead compound discovery and optimization
  ‍
• Predictive modeling for preclinical studies
  ‍
• Adaptive clinical trial design and patient recruitment
  ‍
• Real-world data analysis for post-approval monitoring

The key distinguishing factors of an AI/ML-driven drug are:

• Pervasive Integration: AI/ML is not just an assistive tool; it is embedded across the entire workflow, enabling a more connected and efficient process.
  ‍
• Improved Efficiency: The use of AI/ML leads to tangible reductions in development timelines and costs compared to traditional small molecule or biologic drug programs.
  ‍
• Data-Driven Decision Making: AI/ML models leverage diverse data sources to generate insights that inform critical decisions throughout the lifecycle.
  ‍
• Increased Probability of Success: AI/ML techniques enhance the ability to identify the most promising drug candidates, optimize clinical trial design, and predict real-world performance, improving the likelihood of regulatory approval.

The AI Advantage: Transforming Drug Development

This acceleration is exemplified by Insilico Medicine's INS018_055, the first generative AI-discovered drug to enter Phase II clinical trials3. Insilico Medicine has built a robust AI-powered drug discovery ecosystem comprising three key platforms:

• PandaOmics: Analyzes massive biological datasets to identify and validate potential drug targets.
  ‍
• Chemistry42: Uses generative AI models, including Generative Adversarial Networks (GANs, a type of machine learning) and Transformers (another type of machine learning architecture invented by Google and popularized by OpenAI with ChatGPT), to design and optimize novel molecules.
  ‍
• InClinico: Predicts the likelihood of clinical trial success by leveraging historical clinical data.

By integrating these platforms, Insilico dramatically shortened the drug discovery timeline. Traditional drug development can take 4 - 6 years from target discovery to Phase I trials, whereas INS018_055 reached this milestone in just 30 months, representing a nearly 50% reduction in time and cost. This AI-driven efficiency allowed Insilico to move from initial hypothesis generation to first-in-human trials in record time, setting a new benchmark for AI-powered drug development. The advantages of AI in drug development include:

1. Accelerated Target Identification: AI algorithms analyze genomic, proteomic, and clinical datasets to identify druggable targets with greater precision.
2. Molecular Design & Optimization: Generative AI tools rapidly create and refine novel molecules, predicting their properties before synthesis.
3. Streamlined Clinical Trial Design: AI-powered patient stratification ensures efficient recruitment, reducing trial costs and increasing the probability of success.
4. Post-Approval Monitoring: AI enhances real-world data analysis to track long-term safety and efficacy post-market.

For one and two, we already see many successful examples. For three and four, we still need to wait a few more years. As AI-developed drugs progress through the pipeline, we will better assess AI’s contribution to clinical and post-market phases, even though AI is already being used in these stages independently.

AI in Drug Discovery: Hype vs. Reality

While AI shows enormous potential, it is not a magic wand that ‘invents’ drugs from scratch. Today, its primary impact remains in the early discovery and compound optimization phases.

Reality Check: So far, no AI-developed drug has truly reached the market. Several AI-designed candidates have entered clinical phases, but many have failed, though at least they have failed faster than conventionally developed drugs. Only when we have hundreds of AI-developed drugs successfully approved and commercialized will we be able to truly assess the scientific and economic viability of this approach. And while AI is accelerating drug discovery, we must remain patient, despite its speed, we are still talking about timelines spanning years before we reach this stage.

A fully end-to-end AI drug development pipeline, from target discovery to regulatory approval, remains an aspirational goal rather than today’s reality.

A notable example of this distinction is BXCL501, which has been called an AI-developed drug. However, in reality, it is a repurposed version of Dexmedetomidine (Dex), originally used as an IV sedative. AI was used to identify new indications and optimize formulation, but the molecule itself was neither discovered nor fundamentally developed using AI. This highlights the importance of distinguishing between AI-assisted drug repurposing and true AI-driven drug discovery.

The Road Ahead

According to Jayatunga and his colleagues, AI’s success in Phase I trials (80–90%) suggests it is effective at optimizing drug-like molecules2. AI’s Phase II success rate (~40%) aligns with traditional pharma averages, indicating room for improvement2. The real challenge remains: Can AI-designed drugs sustain high success rates in Phase III and beyond, where the true test of clinical and commercial viability takes place?

The Future of AI-Developed Drugs

The integration of AI/ML in drug development is redefining pharma, but we must be precise in our definitions. Rather than viewing AI as an independent “drug inventor,” we should recognize its role as a powerful accelerator and optimizer within the existing framework of R&D.

As AI-native pipelines expand and regulatory agencies adapt, we may soon see the first fully AI-discovered and developed drugs enter the market. Until then, let’s remain clear-eyed about what AI can, and cannot, do.

Stay tuned for more deep dives in Decoded Drugs.