Nobel Laureate in Chemistry in Dialogue with Shanghai Scientists: As AI Begins Designing Drugs, How Far Are We from "Predicting Life"?

We now have some excellent computational models, such as those developed for electrolytes and proteins; however, we still lack sufficiently robust models tailored to populations and individual patients.

Because every patient is unique. Even in a clinical trial with only 100 patients, there can be significant differences among them, making such predictions extremely difficult.

However, AI is a powerful tool capable of uncovering patterns from data. If, five years from now, we can truly use AI to predict the outcomes of Phase III clinical trials, I would not be particularly surprised. I believe this is a highly realistic and worthwhile question to explore.

Yes. Everyone has unique attributes. While we can measure certain clinical indicators, we still cannot fully understand the mechanisms underlying each individual or explain why such differences arise between individuals.

Future technology may well evolve in this direction. We have already witnessed significant progress in artificial intelligence over the past three years. I believe next-generation AI has the potential to fundamentally transform clinical trials, possibly enabling predictions of Phase III trial outcomes in the future. Should such predictions be validated in actual Phase III trials while maintaining high accuracy, it would represent a highly significant advancement.

In the ideal scenario, our ultimate goal is to truly benefit patients. At the foundation lie targets, molecules, patients, and data. To achieve this, we must continuously repeat a cycle: predict, create, test, learn, and then repeat again, ultimately forming a complete feedback loop.

I believe the cycle itself is more significant than any individual case. Some may feel that artificial intelligence will introduce new obstacles in certain areas, but in reality, we can continuously learn from AI as well. This creates a feedback mechanism—a feedback loop. As we accumulate more cases and medical records, AI can be continually improved and optimized.

Aging is more difficult to address because there is no single molecule that can solve the problem. It may involve numerous factors such as genes, molecules, proteins, metabolism, and the immune system, and these changes occur gradually over a long period, making it highly challenging.

While AI can uncover patterns and identify trends within vast amounts of information, we cannot place all our hopes for extending human healthspan solely on AI, nor should we focus only on achieving a better "score" from a single biomarker. Biomarkers are crucial, but they represent merely the first step, not the ultimate goal.

Recently, we published an article in "Nature BiotechnologyIn this context, proteomics is being employed to assess patient conditions. In the future, AI agents may determine how we apply these techniques. To truly advance research in this field, a critical question remains: How can we effectively leverage AI to bridge micro-level cellular changes with macro-level healthspan? I believe this is an extremely challenging problem.

When discussing aging, it may be necessary to include elderly patients in relevant trials. Such studies are inherently valuable, as long-term observation and testing of older populations can yield more comprehensive data. Different biomarkers of aging and varying dosages may elicit distinct responses to treatment efficacy, influenced by numerous factors. If clinical trial participants also suffer from certain comorbidities, such as pulmonary diseases, treatment protocols would need to be adjusted accordingly. A therapeutic approach effective for one specific demographic may prove ineffective for others. Addressing these disparities will become a critical area of future research. In the future, we may need to explore how to better integrate aging research with clinical trials.

Even when observing healthy populations, we find that there are many signs of aging. We can determine whether an individual is showing signs of aging through various methods. For instance, a slower speed when rising from a chair or cardiac discomfort during treadmill exercise can provide valuable information. To enable related drugs to be genuinely applied to healthy populations, approval from regulatory authorities is still required. I hope that someday in the future, some companies will truly achieve this and transform the development of the entire field.

I believe some truly fundamental changes are underway. We are constantly discovering new patterns and resources, striving to understand them more deeply. The emergence of computational methods has enabled us to gain deeper insights into biology, and we can now even create things that did not previously exist in nature. Our capabilities are becoming extraordinarily powerful, transforming the entire trajectory of scientific progress.

AI will reshape science and also empower it. Yet at its core, science still depends on our ability to ask good questions and on whether surprising experiments occur. AI can accelerate every step—even making mistakes faster. But it also allows us to discover errors more quickly and find the right direction sooner; this is a fundamental logic.

If the cost of developing new drugs can be reduced in the future, we will have the opportunity to create more safe and effective medications. For instance, on social media, you may hear some discussions aboutAnti-aging medicationIn this discussion, while these drugs may help us address population aging, it is essential first to ensure their safety. In the future, we hope to delay the onset of diseases, maintain human physical function as much as possible, and better protect human health.