AI Identifies Drugs for Rare Diseases Treatment

A groundbreaking AI model called TxGNN has been developed, providing hope for the discovery of new treatments for rare diseases – a category encompassing over 7,000 conditions globally that affect around 300 million people.

However, only 5–7% of these illnesses have a medication that has been approved by the FDA; the majority are either undertreated or untreated.

A new artificial intelligence tool has the potential to accelerate the discovery of novel therapeutics from currently available medications, providing hope to patients with uncommon and undertreated ailments as well as the physicians who treat them. The development of new medicines is still a difficult task.

The TxGNN AI model is the first one created expressly to find potential drugs for uncommon illnesses and ailments that have no known cure.

It found potential new drugs from approved medications for over 17,000 illnesses, many of which had no known cure. This is the greatest number of diseases that an AI model has the capacity to handle to date. The model may be used to treat even more disorders than the 17,000 it was tested on in the beginning, according to the researchers.

The work, described Sept. 25 in Nature Medicine, was led by scientists at Harvard Medical School. The technology is freely accessible, and the researchers hope that clinician-scientists will utilize it to find novel treatments, particularly for diseases for which there are no or few current treatments.

“With this tool we aim to identify new therapies across the disease spectrum but when it comes to rare, ultrarare, 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.

“This is precisely where we see the promise of AI in reducing the global disease burden, in finding new uses for existing drugs, which is also a faster and more cost-effective way to develop therapies than designing new drugs from scratch,” added Zitnik, who is an associate faculty member at the Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University.

The two main functions of the new tool are to identify potential treatment choices and their potential adverse effects and to provide an explanation of the decision-making process.

Overall, the tool found potential therapeutic candidates for 17,080 disorders, including those for which there are no known treatments, from roughly 8,000 medications (both FDA-approved and experimental ones currently undergoing clinical trials). Additionally, it forecasted which medications will have side effects and be contraindicated for particular illnesses, something that is primarily determined by trial and error in early safety-focused clinical trials under the current drug discovery approach.

The new technology outperformed the top AI models for drug repurposing by over 50% on average in finding potential new drugs. Additionally, it was 35% more accurate at foretelling which medications will have side effects.

Benefits of taking medications that have already received approval
Because it uses medications that have undergone testing, regulatory approval, and well-understood safety profiles, repurposing existing therapies is an enticing approach to create novel treatments.

The majority of medications include side effects in addition to the original purposes for which they were created and licensed. However, many of these side effects are still unknown and poorly investigated during the early stages of testing, clinical trials, and evaluation, and only become apparent years after millions of people have been using the product. In fact, after receiving initial approval, around 30% of FDA-approved medications have been given at least one further indication for treatment; over time, many of these medications have been given tens or even hundreds of new indications.

This method of repurposing drugs is, at most, haphazard. It depends on patient reports of unanticipated positive side effects or on doctors’ judgment over whether to take a medication—a practice known as off-label use—for a condition for which it was not designed.

“We’ve tended to rely on luck and serendipity rather than on strategy, which limits drug discovery to diseases for which drugs already exist,” Zitnik said.

Zitnik pointed out that the advantages of medication repurposing go beyond ailments for which there is no cure.

“Even for more common diseases with approved treatments, new drugs could offer alternatives with fewer side effects or replace drugs that are ineffective for certain patients,” she said.

Why the new AI tool is superior than current models
Currently, the majority of AI models employed in drug development are trained on one or a small number of diseases. Instead than concentrating on certain illnesses, the new tool was trained so that it could create new predictions using the data that was already available. It accomplishes this by locating characteristics, including common genetic abnormalities, that are shared by a number of disorders.

For instance, the AI model can infer from a well-recognized disease with known treatments to a poorly understood condition with no known treatments by identifying shared disease pathways based on shared genomic foundations.

The research team claimed that this ability makes the AI tool more similar to the kind of reasoning that a human physician may employ to come up with new ideas if they had access to all of the raw data and prior information that the AI model does but that the human brain is not able to access or store.

Large volumes of data, such as DNA information, cell signaling, gene activity levels, clinical notes, and more, were used to train the tool. The model was put to the test and improved by the researchers by having it complete different tasks. Ultimately, 1.2 million patient records were used to confirm the tool’s performance, and it was asked to find potential medication candidates for a range of illnesses.

Additionally, the program was requested by the researchers to forecast which patient features would make the selected drug options inappropriate for a certain patient population.

Asking the tool to locate currently available small compounds that could effectively inhibit the activity of specific proteins implicated in disease-causing pathways and activities was another assignment.

The researchers trained the model to find drugs for three rare conditions: a neurodevelopmental disorder, a connective-tissue disease, and a rare genetic condition that causes water imbalance. The test was intended to assess the model’s capacity for reasoning similar to that of a human clinician.

The suggested medication regimens from the model were then contrasted with what is currently known in medicine regarding the medications’ mechanisms of action. The tool’s suggestions matched the state of medical knowledge in each case.

In addition, the model not only suggested medications for each of the three illnesses, but it also explained its choice. Transparency is made possible by this explanation feature, which also boosts physician confidence.

The researchers issue a warning, stating that further assessment of dosage and delivery schedules would be necessary for any therapy that the model identified. However, they also note that the new AI model would speed up medication repurposing in a way that has never been conceivable before thanks to its extraordinary potential. To assist find potential cures, the team is already working with a number of foundations dedicated to uncommon diseases.

Source Link: Harvard Medical School