Stanford successfully developed AI that can predict drug-related complications

The side effects of drugs - the ones that lead to extremely dangerous complications that can be life-threatening are really hard to predict because according to estimates, there are more than 125 billion potential complications. 'It is impossible to test a new drug by combining it with available drugs, because it takes up to 5,000 different experiments,' said Marinka Zitnik, a graduate student in machine science. Calculated at Stanford University (USA), said. 'In fact, we don't know what will happen'.

To solve this problem, Zitnik and Jure Leskovec, a professor of the same branch, developed Decagon - a system with the ability to predict possible side effects when drugs are used together. Sponsored by the National Science Foundation, the National Institutes of Health, Agency for Advanced Defense Research Projects, Stanford Data Science Initiative, and Chan Zuckerberg Biohub (funds established by Facebook CEO Mark Zuckerberg and his wife Priscilla Chan), the system has digitized more than 19,000 proteins in the human body capable of interacting with each other as well as interacting with drugs.

After synthesizing more than 4 million drug, protein and side effects, the team then used a "deep learning" algorithm to make predictions which combination would lead to complications and What earned was far more promising than expected.

Marinka Zitnik, graduate student in computer science at Stanford University - one of the authors of the study.

Through testing, Decagon has the ability to accurately predict side effects due to combining drugs with an accuracy of up to 69%. For example, the predictable system when taking atorvastatin (cholesterol-lowering drug) in conjunction with amlodipine antihypertensive drugs may lead to myositis. Furthermore, this AI can make its judgments about the possibility of complications with results that coincide with the 10 new side effects recently confirmed by medical researchers.

At the present time, Decagon still has a drawback: it is only possible to make predictions about possible side effects between pairs of drugs but not more. And this is being worked on by scientists to improve in an updated version that will be released in the future.

Known in the medical industry today, pharmaceuticals are not the only area that is becoming more and more progressive thanks to Artificial Intelligence. Researchers are also using "machine learning" in predicting disabilities in newborns, in addition to many developing startups AI to serve to improve human health, Google is also As a candidate in this race with DeepMind Health, the project aims to develop a suite of applications that can help doctors determine the risk of complications in patients.

Analysts at McKinsey market consulting firm estimate that algorithms like this can help pharmaceutical companies save up to $ 100 billion per year.