Artificial intelligence to find new uses for drugs in record time

When a machine learning algorithm (a form of artificial intelligence) is used to predict which diseases a drug that was designed for other uses might be used (a repurposing known as “repurposing”), the algorithm may recommend some drugs, but not explains why, and this raises doubts regarding the reliability of the prediction. Therefore, it would be ideal to have a repositioning mechanism that also explains why it predicts the way it does.

This is where a group of researchers from the Polytechnic University of Madrid (UPM) in Spain has intervened by providing a solution. The team has just developed a drug repositioning method with a clear emphasis on interpretability, as it aims for the system to offer explanations for why it proposes treating disease X with drug Y.

This new algorithm – called “XG4REPO” (eXplainable Graphs for Repurposing) – not only repositions, but also presents the results in a way that is understandable, indicating which biological mechanisms are used for the prediction. This allows your predictions to be validated by medical experts, who can immediately assess whether it is a valid explanation or not, thus generating much more robust predictions and saving time that would otherwise be spent searching for explanations.

The process of creating medicines is slow and expensive, as it involves numerous tests to ensure the safety of the new medicine in order to obtain authorization for its marketing by health authorities. An alternative that is increasingly gaining strength to alleviate this situation is the repositioning of existing medications, which consists of identifying new applications for drugs that are already approved. This means using a drug that already exists to treat a different disease than the one that was in mind when the drug was designed.

Medication repositioning, as a technique, has a series of advantages that cannot be ignored. The first is that it allows drug development times to be significantly shortened, since the drug is approved and its side effects are known. The second great advantage is related to the cost of developing the drug, which does not require repeating expensive safety tests, first in animals, and then in clinical trials. Likewise, most drugs developed in the laboratory do not reach the market due to their adverse effects, a problem that repositioning does not have.

These advantages allow repositioning to be considered a technique that can introduce major changes in medicine. On the one hand, it allows us to develop treatments for new diseases much more quickly than creating a medicine from scratch. During the COVID pandemic, for example, repositioning made the news for attempts to use different medications to treat this new disease. But, in addition, repositioning is a great hope for patients with rare diseases, since it would allow the development of treatments at a low cost for the laboratory.

Symbolic artistic recreation of the concept of rapid search for new uses of drugs using artificial intelligence. (Illustration: Amazings/NCYT)

At a technical level, a medication affects a specific biological process; For example, paracetamol blocks part of the human body’s pain impulse. The great challenge of repositioning is to identify which patterns, affected by a specific drug, appear in other diseases. Thus, if two diseases have a similar pattern, and a certain medication is used to treat the first, it is likely that it can also be used for the second. But performing this pattern identification by experts is an expensive process that requires extensive knowledge of diseases and their mechanisms. On the other hand, the machine learning techniques currently available are very good at this pattern detection, hence the great recent interest that has arisen in the repositioning of medications using artificial intelligence.

However, artificial intelligence techniques present the problem of interpretability. In an attempt to solve it, the team led by Ana Jiménez, from the Polytechnic University of Madrid (UPM) has found the solution to the problem described thanks to the design of a new algorithm that she and her colleagues have called “XG4Repo”. This constitutes a framework for drug repurposing using knowledge graphs that predict diseases that can be treated with a given compound. Along with the prediction, the model provides the rules that support the prediction and the importance of the rule.

To demonstrate the effectiveness of “XG4REPO”, the researchers tried to predict the application of three known anti-cancer drugs and detected that, among the algorithm’s predictions, there were many that were already in the initial clinical trial phase. This means that there is medical evidence that validates the predictions made by “XG4REPO”. Therefore, as Professor Santiago Zazo, who has been part of the work team, points out, “this mechanism constitutes one more step towards the application of artificial intelligence techniques in the medical field, not to replace experts, but to facilitate their analysis.” of a large amount of data in a short time, and accelerate the drug development process.”

Ana Jiménez and her colleagues present the technical details of their new algorithm in the academic journal Scientific Reports, under the title “Explainable drug repurposing via path based knowledge graph completion.” (Source: UPM)