Enhancing AI's Role in Identifying Therapeutic Goals
In a recent panel discussion, experts from academia and pharmaceutical industries gathered to discuss the potential of artificial intelligence (AI) in drug discovery. The panel, which included Manoj Kandpal from the Rockefeller University, Michael Steinbaugh from Merck, Shameer Khader from Sanofi, Simona Cristea from the Dana-Farber Cancer Institute, and Bissan Al-Lazikani from the University of Texas MD Anderson Cancer Center, explored the transformative impact AI can have on drug target predictions, especially when combined with multi-omics and proteomics data.
AI is already showing promise in speeding up key steps in the drug discovery process. When paired with rich datasets, high-quality biological models, and human expertise, AI is helping to accelerate the process significantly. However, the experts emphasized that the success of AI-driven target discovery depends more on building high-quality, diverse datasets than developing sophisticated models.
One of the challenges in drug discovery is the discrepancy between drugs showing promise in current models and their failure in humans. As Bissan Al-Lazikani pointed out, "We all talk about the nightmare of 95% of cancer drugs entering clinical trials failing. Well, 100% of those succeeded in mice." AI, when used correctly, could help bridge this gap by providing more accurate predictions.
A paper published recently showed that proteomics data was far superior to any other individual modality at predicting patient drug response. This suggests that the integration of proteomics data into AI systems could lead to more effective drug discovery.
While AI holds great potential, it is not a replacement for human expertise. The panelists stressed the importance of pairing AI with experts who can interpret results and guide decisions in AI-driven research. Human judgement is essential in determining which targets are worth pursuing and which aren't, even in contexts such as target identification.
The panel also highlighted the need for better biological models to address the challenge of proving that drugs work in humans, which remains a bottleneck in the drug discovery process. Without real-world data, simulations may mislead scientists rather than accelerate discovery.
The panelists also cautioned against over-reliance on artificial or generated inputs in AI-driven target discovery. AI decay can occur when AI systems are trained on synthetic data that they generate and iterate. To combat this, efforts should be made to generate meaningful biological data.
Despite these challenges, the future of AI in drug discovery holds promise for bringing new treatments to patients. The panelists agreed that with the right approach, AI could revolutionize the drug discovery process, making it faster, more efficient, and more effective.
