The artificial intelligence model can predict the side effects of new combination therapies

According to the results presented at the AACR Annual Meeting 2022 on April 8-13, basic data from the artificial intelligence model can predict the side effects of new combination therapies.

Physicians are challenged by a real-world problem where new combination therapies can lead to unpredictable results. Our approach will help us understand the relationship between the effects of different drugs in relation to the context of the disease. “

Bart Westerman, PhD, senior author of the study and associate professor at the Cancer Center Amsterdam

Many types of cancer are increasingly being treated with combination therapies, in which physicians seek to increase efficiency and reduce the chances of treatment resistance. However, such combination therapies can add multiple medications at once to a patient’s already complex list of medications. Clinical trials testing new drugs or combinations rarely result in other drugs being taken outside of the patient-tested treatment regimen.

“Patients seeking treatment typically use four to six medications per day, making it difficult to determine whether a new combination therapy is harmful to their health,” Westerman said. “It is difficult to assess whether the positive effect of combination therapy justifies its negative side effects on a particular patient.”

Westerman and colleagues, including Aslı Küçükosmanoğlu, a graduate student who presented the study, called for the use of machine learning to better assess adverse events resulting from new drug combinations. They collected data from the US Food and Drug Administration’s Advertising Event Reporting System (FAERS), which has a record of over 15 million adverse events. Using a method called dimensional reduction, they grouped frequently co-occurring events to facilitate analysis and strengthen the links between the drug and its adverse profile.

The researchers then presented the data into a convolutional neural network algorithm, a type of machine learning that simulates the way human brains make connections between data. Adverse events for individual treatments were used to train the algorithm, which identified common patterns between drugs and their side effects. The detected patterns are encoded in what is known as the “hidden space”, which simplifies the calculations by representing each negative event profile as a string of 225 numbers between 0 and 1, which is then decoded back to the original profile.

To test their design, the researchers provided their model with invisible negative event profiles of combination therapies, called an “atlas of adverse events”, to see if they could identify these new profiles and decode them correctly using hidden space descriptors. This showed that the model could detect these new patterns, proving that the measured combined profiles could be reversed to those in each drug in combination therapy.

The negative effects of combination therapy can be easily assessed, says Westerman. “By simple algebraic calculation of latent space descriptors we were able to determine the sum of the individual treatment effects,” he explained. “Because the algorithm is trained to detect global patterns, this approach reduces noise in the data so that it can accurately capture the side effects of combination therapies.”

Westerman and colleagues further validated their model by comparing the expected negative event profiles of combination therapies with those observed in the clinic. Using data from the FAERS and US Clinical Trials database, researchers have shown that the model can accurately retrieve negative event profiles for some commonly used combination therapies.

The complexity of combination therapies is the new, potentially unpredictable side effects that can arise when drugs are combined. Using the additive samples identified by the model, the researchers were able to distinguish the additive side effects from the synergistic side effects of the drug combinations. Westermann says it helps them better understand what happens when complex negative event profiles are linked to each other.

Researchers are developing a statistical method to calculate the accuracy of their model. “Because the landscape of drug interactions is so complex and involves so many molecular, macrophage, cellular, and organ processes, our approach is unlikely to lead to black and white conclusions,” Westermann said. “Adverse events Atlas is still in the proof-of-concept phase, but the most important discovery is that we were able to obtain snapshots of drugs, diseases, and the interaction of the human body as described by millions of patients.”

The limitations of this study are the potential difficulties in comparing this data with even less data, as well as the limited application of the model to clinical practice until further validation is provided.


American Association for Cancer Research

Leave a Comment