Machine learning has a plethora of applications. One area in which machine learning is being used is the field of medicine. Researchers perform experiments in labs to study effects of medicine on different animals. When a drug works on a monkey or mice, they start a clinical trial on humans (after endless amounts of paperwork). If the clinical trial is successful on a diverse sample of subjects, then the FDA approval starts. After FDA approval, the drug can be marketed for mass distribution across the country. At this point, the drug makers spend money on marketing, sales representatives, providing incentives to physicians that agree to prescribe the drug, and advertising on TV and the internet. The drug is known to treat a certain condition, and has a list of known side effects. Millions of people across the country take this drug and it works for most of them. This is a system that has been in use for decades.
This seems to be working fine. So where is the problem? The problem lies with studying effects of multivariate medicine. A lot of people consume more than one medicine at a given time. They can be on multiple prescriptions. We study the effects of a single medicine to understand what it does to the body (its cure and side effects). However, when you consume two or more medicines together, you have a multivariate problem. The equation contains two or more variables, each having an effect, but together with other medicine, having a different effect. With each medicine you add (another variable), the problem gets harder. Medicine A can have some effects on a person. Medicine A + Medicine B can have a different effect. But when you mix Medicine C, the results change. You can have endless permutations and combinations. While you can study the effects of this on mice in labs, it would take an en eternity as you would need a sample of mice on Medicine A + B, a sample on Medicine A + C, a sample on Medicine B + C, a sample on A + B + C, etc. This would become exponentially harder when you add each variable.
So how do we solve this problem? In the last few decades, we did not solve this problem. We focused on researching new drugs, doing trials, getting FDA approval, and marketing. However, with new breakthroughs in Data Mining and Machine Learning, we can use computers to predict the outcome of these combinations with a decent confidence level. We will not need an eternity and endless mice, we just need some computers in the cloud crunching numbers. Why do we care about this now? It’s because this will help us understand effects of medicine taken together. So a patient can go to a computer, enter the list of medications, and know the real list of side effects that the patent faces when that combination of medicine is taken together. Some medicine taken together can have lethal effects even though taken separately, they are safe. We do know that some medicines don’t go together. But we don’t have a complete list of every combination of drugs that are FDA approved. Furthermore, combining some drugs can have interesting effects. They can be the cure for a disease that we did not have a cure for. So by using Machine Learning to understand side effects for the machine, we will also discover combinations that would be cures for diseases that we currently don’t have a cure for. We will be able to use the knowledge of medicine that we have accumulated by lab experiments, and generate exponentially more knowledge about medicine by just computers crunching numbers in the cloud.
Machine Learning is being used in a lot of fields for different purposes, and now we will be able to accelerate discoveries in the field of medicine and learn about side effects that we did not know about.