For many American citizens, taking prescription pills is part of a daily routine. About seven out of ten people get at least one drug that can be made up of several pills a day, according to the Mayo Clinic, and 15% get five or more. The problem is that it is considered quite difficult to predict the side effects that may result from taking multiple pills on a daily basis and let alone the combination of different kinds, which are in fact some considered to be narcotics.
According to scientific studies, there are more than 125 billion possible complications, many of which are not dangerous, but some are even considered deadly.
“It is virtually impossible to test a new drug in conjunction with all other medicines and not to have some minor complications, and the worst of all is that we do not know what can result from this result,” said Marinka Zitnik, a scientist at Computer Science at Stanford University.
To address the problem, Zitnik and Jure Leskovec, an associate professor of computer science, developed an AI system called Decagon – and can predict possible side effects for drug combinations. They described their research in a paper titled Modeling polypharmacy side effects with graph convolutional networks that was presented this week at the Conference of the International Society of Computational Biology 2018 in Chicago.
The system, funded in part by the National Science Foundation, the National Institutes of Health, the Advanced Defense Research Program, Stanford Data Science Initiative and Chan Zuckerberg Biohub, has the ability to model more than 19,000 proteins in the human body who interact with each other and with drugs. Using a deep learning algorithm that understands about four million known associations between drug and protein side effects, it is even able to make various predictions and accurately infer standards that developers have set for that purpose.
Decagon accurately predicted the side effects of new drug combinations, far exceeding the standard methodology used in this field. For example, he predicted that the drug for cholesterol with the active substance atorvastatin, when taken with medication including amlodipine for blood pressure, could lead to muscle inflammation. In another case, he accurately predicted ten side effects recently confirmed by medical researchers.
“It was particularly surprising that the protein interaction networks reveal so much about the side effects of drugs,” Leskovec said.
Currently, the system is accompanied by some limitations – and at this stage it only looks at the side effects associated with the combination of medicines. But in the future, the team hopes to create an improved version that can handle more complex shapes.
“Today, the side-effects of drugs that some drugs are also found to be completely random, and our approach has the potential to lead to more effective and safer healthcare,” said Leskovec. Pharmacology is not the only area where healthcare can benefit from artificial AI intelligence. Researchers use mechanical learning to predict possible future disabilities in infants.
Also, some private companies operating in this area have developed equally important systems, such as Clew Medical offering an AI-based analysis platform that prevents life-threatening complications for patients. Google has also invested significant financial resources into artificial intelligence-based healthcare programs such as DeepMind Health. Analysts believe that engineering learning algorithms could help medical and pharmaceutical companies save up to $ 100 billion a year from unnecessary costs.