Innovative approaches to drug discovery are needed to help find effective treatments for the myriad of diseases that afflict humanity. The process is extremely lengthy and expensive, with timelines up to 15 years, and costs over a billion dollars. Now TxGNN, a groundbreaking artificial intelligence (AI) model has been developed which promises to improve the way opportunities are identified for diseases with limited treatment options and made available for free to encourage clinician-scientists advance the search for new therapies.
Leveraging recent advances in geometric deep learning and human-centered AI, Harvard’s Zitnik lab has introduced TxGNN, a model designed to innovate drug repurposing and discovery. TxGNN is a graph neural network pre-trained on a comprehensive knowledge graph encompassing 17,080 clinically-recognized diseases and 7,957 therapeutic candidates. This extensive dataset allows TxGNN to process various therapeutic tasks, such as predicting indications (suitable uses for a drug) and contraindications (situations where a drug should not be used), in a transparent format that makes drug-hunting easier for researchers.
TxGNN is the first AI model specifically developed to identify drug candidates for rare diseases and conditions with no treatments and represents the largest number of diseases that any single AI model can handle to date. The researchers believe that TxGNN could be applied to even more diseases.
Zero-Shot Inference: A Game Changer
One of the most remarkable features of TxGNN is its ability to perform zero-shot inference. This means that the model can make predictions about new diseases without requiring additional parameters or fine-tuning on ground truth labels. In other words, TxGNN can identify potential therapeutic uses for diseases it has never encountered before, making it a powerful tool in the fight against rare and neglected diseases. According to experts, zero-shot inference is a technique where a model can classify or predict information for a completely new category, even if it hasn’t seen any training examples for that specific category.
Addressing Rare and Undiagnosed Diseases
There are more than 7,000 rare and undiagnosed diseases globally. Although each condition affects a small number of individuals, collectively these diseases impact approximately 300 million people worldwide, exerting a staggering human and economic toll. With only 5 to 7 percent of these conditions having an FDA-approved drug, they remain largely untreated or undertreated. Developing new medicines for these diseases is a daunting challenge, but TxGNN offers hope and the free tool was released specifically to facilitate more work in the field.
The performance of TxGNN has been nothing short of impressive. In addition to matching drugs for specific conditions, it predicted which drugs would have side effects and contraindications, something that the current drug discovery approach identifies mostly by trial and error during early clinical trials focused on safety. In evaluations, the model demonstrated significant improvements over existing methods, achieving up to 49.2% higher prediction accuracy for indications and 35.1% higher accuracy to identify contraindications.
Historically, drug repurposing—identifying new therapeutic uses for approved drugs—has often been a matter of serendipity. By utilizing a geometric deep learning framework, TxGNN can make therapeutic predictions even for diseases with no existing medicines. This capability is particularly valuable for addressing the needs of complex, neglected, or rare diseases, which often lack pre-existing indications and known molecular target interactions.
Interpretable and Transparent Predictions
One of the key strengths of TxGNN is its interpretability. The model’s Explainer module offers transparent insights into the multi-hop paths that form TxGNN’s predictive rationale. This feature allows clinicians and scientists to contextualize and validate the model’s predictions, making it easier to trust and act upon the suggested therapeutic candidates. During user studies, medical experts found these explanations instrumental in understanding and validating TxGNN’s predictions.
TxGNN pinpoints shared disease mechanisms based on common genomic underpinnings, allowing it to extrapolate from a well-understood disease with known treatments to a poorly understood one with no treatments. This capacity brings the AI tool closer to the type of reasoning a human clinician might use to generate novel ideas if they had access to all the preexisting knowledge and raw data that the AI model does but that the human brain cannot possibly access or store.
The tool was trained on vast amounts of data, including DNA information, cell signaling, levels of gene activity, clinical notes, and more. The researchers tested and refined the model by asking it to perform various tasks. The tool’s performance was validated on 1.2 million patient records, identifying drug candidates for various diseases and predicting patient characteristics that would render the identified drug candidates contraindicated for certain populations.
Future Directions
The success of TxGNN opens up exciting new possibilities for drug discovery. The model’s ability to generalize to diseases with few treatment options and perform zero-shot inference makes it a versatile tool that can be adapted for other use cases, such as drug target discovery and targeted therapy selection. Moreover, the multi-disease predictive strategy employed by TxGNN suggests that comprehensive approaches to drug repurposing can yield more repositioned drug candidates than traditional single-area approaches.
TxGNN represents a significant advancement in the field of drug discovery. By leveraging the power of geometric deep learning and human-centered AI, the model offers a promising solution to the challenge of identifying new therapeutic opportunities for diseases with limited treatment options. With its impressive performance, real-world validation, and transparent predictions, TxGNN is poised to make a lasting impact on the way we discover and repurpose drugs, ultimately improving patient care and outcomes.
