Precision medicine is reshaping healthcare by moving beyond the traditional one-size-fits-all model. Instead of treating every patient the same, it considers the unique characteristics of each individual, such as genetic makeup, lifestyle and environment, to guide more targeted diagnosis and treatment.
For example, two patients with the same cancer may receive different treatment plans based on the mutations in their tumors or how their bodies are likely to respond to different drugs based on their genetic profile. Over the past few years, Generative AI (GenAI) has emerged as a powerful catalyst in this field. It can help predict drug response and adverse events in patients, overcome data limitations by generating synthetic data, enhance clinical efficacy, and accelerate discovery, bringing more personalised therapies to patients faster than ever before.
Predicting drug responses with generative AI
GenAI is enabling better predictions as to how individual patients are likely to respond to specific treatments as well as identifying potentially adverse events before they happen. By training on biomedical data like genomic, transcriptomic, and clinical records, these models can assess a drug’s likely efficacy and toxicity based on a patient’s unique profile. For example, a 2023 study by Wang and colleagues leveraged Multi-Omics Integrated Collective Variational Autoencoders (MOICVAE), an AI model capable of accurately predicting drug sensitivity for 25 drugs across seven kinds of cancer. In a separate study, Shi and colleagues proposed CSAM-GAN, a generative adversarial network based on sequential channel-spatial attention modules able to predict patient prognosis in lower-grade glioma and kidney renal clear cell carcinoma.
Models like these are helping clinicians match the right therapies to the right patients, while minimizing risks through earlier, data-driven insight.
Overcoming data limitations with single-cell AI models
The rapid growth of single-cell sequencing data is creating new opportunities. So-called single-cell foundation models based on generative pre-trained transformers are able to distill biological information about genes and cells and be fine-tuned for biomedical tasks, like cell identification, even when data is incomplete. Their broad cellular knowledge makes them highly generalisable across a range of biomedical applications. Relatedly, GenAI can also be used to create synthetic biomedical training data that closely mimics the statistical variation found in real world patient data. In the case of rare diseases with few patients, for example, researchers can train models on thousands of AI-generated examples that match the underlying condition to create more powerful and accurate models. Additionally, these datasets can be stripped of personally identifying information and are thus suitable for publication without fear of violating patient confidentiality.
Enhancing clinical development
GenAI offers new ways to identify patient subgroups and the relevant biomarkers, crucial for clinical development and maximising drug efficacy. In cancer and other complex diseases, different patients frequently respond differently to the same therapy. Uncovering which biomarkers (or combination of biomarkers) differentiate treatment responders from non-responders is a key part of precision models.
GenAI excels at finding patterns across complex datasets that are hard to analyse using more traditional statistical methods. CSAM-GAN, for instance, is able to integrate a patient’s DNA profile, RNA profile, and histopathology imagesto predict outcomes — and crucially, pinpoint which biomarkers drive those predictions, whether a gene, pathway, or tissue feature.
Looking ahead, GenAI could even enable the creation of “digital twins” for each patient. Every available treatment plan could then be simulated with the digital twin to identify how each one is likely to play out. While still an emerging concept, this could one day transform personalised medical care.
Accelerating drug discovery
Perhaps most strikingly, GenAI is accelerating the traditionally slow and costly process of new drug development. Bringing a new drug to market can take well over a decade and costs billions. GenAI has the potential to dramatically cut both timelines and expenses.
GenAI is streamlining and automating stages that traditionally would take researchers months or years. Instead of physically synthesising and testing thousands of possible drugs, GenAI models can generate novel drugs that have a high likelihood of meeting a specific criterion. For example, these models can identify compounds that have a high chance of binding to a particular receptor (say, serotonin receptors) while having a low chance of toxicity. This narrows down the list to the most promising candidates, reducing early-stage development from years to mere weeks or months.
GenAI and the future of precision medicine
Generative AI is rapidly becoming a cornerstone of precision medicine. Already, it is helping to identify individual drug responses, uncover biomarkers, generate synthetic datasets, and accelerate drug discovery. And this is only the beginning – these models will become even more capable and impactful as they continue to evolve. At its core, generative AI is, making medicine more predictive, efficient, and personalised. It’s not difficult to imagine a future where a doctor consults an AI colleague that runs virtual trials of every available therapy based on a patient’s biomedical data. With this level of insight, clinicians will be able to make faster, more accurate decisions – delivering the right treatment to the right patient at the right time.
