An artificial intelligence model developed by researchers at the Yale School of Medicine and Google has identified a novel approach to cancer treatment using the drug Silmitasertib. This drug, traditionally recognized for its role in inhibiting cancer growth, was identified by the AI model as a means to enhance the immune system’s ability to detect tumors. This groundbreaking discovery was the result of a collaboration between Yale’s team, led by Professor David van Dijk, and Google’s leading AI labs, Google DeepMind and Google Research.
The AI model, known as Cell2Sentence, suggested that Silmitasertib could increase antigen presentation, a critical process that enables the immune system to identify and target cancerous cells. Remarkably, this hypothesis was not based on any prior published research, making the findings all the more surprising. The lab conducted experiments on human skin and pulmonary cells, validating the model’s prediction, as detailed in a preprint paper published in October 2024.
Innovative Research Methodology
The team at Yale has been exploring a new method for understanding human cells through single-cell RNA sequencing. This approach focuses on measuring gene expression at an individual cell level. According to Syed Rizvi, a graduate student involved in the study, understanding genetic patterns within cells is vital for distinguishing between healthy and malignant cells.
To decode this complex data, the researchers employed natural language processing, a branch of AI designed to interpret human language. By transforming numerical gene expression data into sentence-like structures, the AI model was able to recognize biological patterns crucial for identifying cell types. For instance, a simple cellular sentence might include sequences of genes that characterize specific cell types.
Despite early successes, the initial model used, based on GPT-2, faced limitations due to its smaller scale of 774 million parameters. In contrast, GPT-4, released in March 2023, boasts 1.76 trillion parameters, significantly enhancing the model’s complexity and capability. The challenges posed by the constrained computing resources at Yale restricted the team’s ability to utilize the more advanced models available.
Collaboration with Google
In 2024, an AI workshop hosted by Google at Yale marked the beginning of a pivotal collaboration. The partnership allowed Yale researchers to access Google’s extensive computing resources, which are among the largest in the world. According to Shekoofeh Azizi from Google DeepMind, this collaboration evolved organically as the two teams recognized their complementary strengths.
With Google’s support, the researchers transitioned from GPT-2 to Gemma-2, a more sophisticated AI model that expanded Cell2Sentence to 27 billion parameters. This advancement enabled the team to conduct more complex analyses, including predicting how drugs could impact human cells.
The breakthrough came when the AI model was tasked with identifying drugs that could enhance immune signals in the presence of sickness markers. The model successfully identified Silmitasertib, leading to experimental validation that confirmed the drug’s potential to influence immune response. Azizi noted that this success exemplifies the capability of large-language models to engage in complex biological reasoning.
Looking ahead, the collaboration with Google signals a transformative shift in drug development. A study from 2024 indicated that the average cost to develop a new drug can reach approximately $500 million, with pharmaceutical companies often investing billions in research and development. Van Dijk expressed optimism that their findings could shorten the pre-clinical phase of drug development by guiding researchers toward experiments with a higher likelihood of success.
As the team continues to refine their AI models, Van Dijk envisions a future where AI can simulate the human body in its entirety, enabling faster and safer drug testing. “Imagine a virtual human that can simulate everything biologically about real humans,” he said. “This could significantly improve our ability to identify effective and safe treatments.”
The research team’s findings underscore the potential of AI to revolutionize cancer treatment and drug discovery, marking a significant milestone in the integration of technology and medicine.