Across the globe, more than 7,000 rare and undiagnosed diseases have been identified. Although each condition affects fewer than 1 in 2,000 individuals, their collective impact is significant, affecting around 300 million people worldwide.
The toll on human lives and the economy is immense, as most patients endure prolonged suffering without effective solutions. Despite the urgent need, only 5 to 7 percent of these diseases have FDA-approved treatments.
This leaves the majority of conditions without proper medical interventions, creating significant gaps in care and limiting treatment options for millions. The lack of widespread treatment solutions underscores the need for better drugs and more effective treatments for rare diseases.
Researchers from Harvard Medical School have developed a new artificial intelligence (AI) tool that can accelerate the discovery of new therapies from existing medicine. This “repurposing” of drugs offers new hope for patients suffering from rare diseases.
The new AI model, named TxGNN, uses a unique graph neural network (GNN) approach to identify potential drug candidates by analyzing shared genomic and biological features across diseases. TxGNN is trained on a medical knowledge graph, which includes extensive biomedical data.
Researchers who developed TxGNN assert that it is the first model of its kind, with the potential to play a critical role in discovering drug candidates for rare diseases and conditions that currently lack effective treatments.
During the initial experiments, TxGNN successfully identified nearly 8,000 existing medicines for over 17,000 diseases, many of which currently lack any treatments. This achievement marks the largest number of diseases that any single AI model has been able to address to date. The Harvard team is confident that this model could perform even better when applied to more diseases.
TxGNN has been made available for free to encourage scientists around the globe to utilize it for their research for new drugs and treatments.
“With this tool, we aim to identify new therapies across the disease spectrum but when it comes to rare, ultra rare, and neglected conditions, we foresee this model could help close, or at least narrow, a gap that creates serious health disparities,” said lead researcher Marinka Zitnik, assistant professor of biomedical informatics in the Blavatnik Institute at HMS.
Repurposing drugs offers key benefits compared to developing new treatments due to their established safety profiles and regulatory approval. Additionally, since these drugs have already undergone clinical trials, the time required to bring them to market is typically much shorter, allowing for quicker patient access.
The research, detailed in Nature Medicine, highlights TxGNN’s innovative use of zero-shot predictions, which significantly enhance both accuracy and interpretability, offering a groundbreaking approach to drug repurposing for rare and untreated diseases.
Most traditional AI models for drug repurposing are trained on specific datasets to make predictions for particular diseases. However, TxGNN’s zero-shot learning capability enables it to be applied to diseases with no prior data, broadening its applicability to a much wider range of conditions without the need for extensive model retraining on each specific disease.
A key component of the TxGNN model is its Explainer module, which enhances transparency by revealing how the AI generates its drug predictions through multi-hop connections within a medical knowledge graph. This feature enables experts to understand the reasoning behind the model's recommendations, making TxGNN's decisions more interpretable and trustworthy for clinical applications.
AI and machine learning (ML) are revolutionizing the pharmaceutical industry by significantly enhancing the traditional drug discovery and development process.
For instance, Insilico Medicine and the University of Cambridge developed a new AI-based technique that has enabled a breakthrough in identifying new targets for Alzheimer's and other diseases with protein phase separation (PPS). Additionally, OpenFold has harnessed the power of AI to create a new program capable of predicting the structure and interactions of biological molecules with unparalleled accuracy.
The possibility of repurposing drugs presents a strategic approach to overcoming challenges in drug discovery, such as long development times and high costs. The research team is already collaborating with several rare disease foundations to help identify possible treatments.
However, the team emphasizes that to maximize the effectiveness of the model for complex and neglected diseases, further validation through larger studies is essential to address potential biases and enhance its overall impact.