Summary of Key Points
Recently, despite some highlights in AI-driven pharmaceutical development (with multiple pipelines entering Phase III clinical trials), the industry and investors have become more cautious: The speed brought by AI does not necessarily equate to cost savings. The ultimate value of a drug depends on its clinical efficacy, not just on whether it uses AI technology. Currently, there is an overemphasis on chemical models (molecular design) at the expense of biological models (understanding disease mechanisms), leading to low success rates in clinical trials. There is a need to shift from the "First-in-Class (FIC)" paradigm to the "First-in-Disease-Domain (FID)" approach—starting with the actual needs of patients and integrating biological and chemical models to improve clinical success rates from the outset.
Detailed Analysis
1. AI Pharmaceuticals: Coolness Behind the Highlights
In the past six months, there have indeed been several notable successes in AI pharmaceuticals, with multiple AI-designed drugs entering Phase III trials (such as those developed by Insilico Medicine for idiopathic pulmonary fibrosis and DeepMind for small molecule therapies). However, the industry is no longer overly optimistic for two main reasons:
- Speed does not equate to cost savings: While AI can quickly identify potential candidates, any time and cost savings in the early stages may be offset by the costly and time-consuming clinical validation process.
- The importance of clinical efficacy: The real value of a drug lies in its ability to address unmet patient needs, not just in whether it uses AI technology.
In other words, AI is a tool; only drugs that can effectively treat diseases are truly valuable.
2. The “Valley of Death” in Clinical Trials
New drug development faces a significant challenge known as the “Valley of Death,” where only about 25.6% of candidate drugs succeed in Phase II trials. AI is particularly powerless at this stage because it can only generate hypotheses based on existing research, but verification requires experiments on animals and humans, which are both expensive and time-consuming.
- AI’s limitations: AI cannot verify these hypotheses; it can only suggest potential candidates for further testing.
3. The Imbalance in Investment
There is a significant imbalance in investment between chemical and biological models in AI pharmaceuticals:
- Chemical Models (Designing the “Key”): These focus on molecular design, which is currently the most widely used area of AI applications.
- Biological Models (Finding the “Lock”): These aim to understand disease mechanisms and identify target molecules that are relevant to the disease.
This imbalance means that even if chemical models produce promising results, they may be ineffective if the wrong targets are targeted.
4. The FID Paradigm: Moving from “Drug-Finding-Disease” to “Disease-Finding-Drug”
The industry is proposing the “First-in-Disease-Domain (FID)” approach as a better alternative to the traditional “First-in-Class (FIC)” paradigm:
- FIC: This starts with identifying new molecules and then determining their potential therapeutic applications.
- FID: It begins by understanding unmet patient needs, using biological models to identify target molecules, and then designing drugs accordingly.
For example, Dr. Zhao Yu’s team used biological models to understand the mechanisms of pancreatic cancer and successfully designed a new drug that entered Phase I clinical trials, increasing the chances of success from the outset.
5. The Way Forward: Virtual Clinics and Policy Support
To overcome these challenges, two approaches are needed:
- Technological Solutions: Virtual clinics using AI to simulate patient responses can reduce the need for animal and human trials, lowering costs and risks.
- Policy Support: More funding and resources should be allocated to biological models, which require extensive data and computational power. This may involve government initiatives or large-scale collaborations to establish clear funding guidelines and long-term evaluation frameworks.
Only by addressing these issues can AI pharmaceuticals truly overcome the “Valley of Death.”
Conclusion
AI in pharmaceutical development is not a disruptor but an enabler. Its future lies in more accurately addressing patient needs through the integration of biological and chemical models, thereby improving clinical success rates and becoming a key driver of drug innovation.