MIT researchers used machine learning to improve the efficiency of drugs in the body
When a disease is studied, there are several factors that need to be addressed. The origin, sources, causes, effects, etc. form the preliminary part of the study. There’s yet another side of the story – the drug aspect of the disease. Study on the same revolves around – which organs of the body do the drug enter, how much of the drug amount gets absorbed, how does the body reacts to it, and so on. When talking about what the body does to a drug, there are nanoparticles out there that can improve the whole process. These nanoparticles comprise lipids, polymers, and in some cases both. On the flip side, the process isn’t easy.
Also, combinations of small-molecule cancer drugs with two small-molecule dyes showcase the ability to get self-assembled with these nanoparticles. This is exactly where the real challenge creeps in. How to predict which of those small molecules assembled with the nanoparticles? What poses an obstacle is that there could be over millions of pairings possible. This issue transformed into a subject of interest for MIT researchers. They developed a platform using machine learning that’d identify self-assembling nanoparticles quickly.
Some combinations of and have been shown to self-assemble into nanoparticles with extremely high payloads of drugs, but it is difficult to predict which small-molecule partners will form nanoparticles among the millions of possible pairings. On this screening platform, these researchers made headlines by screening 2.1 million pairings of small-molecule drugs along with inactive drug ingredients. They successfully identified 100 new nanoparticles. It doesn’t stop here. This discovery holds potential in the treatment of diseases like malaria, asthma, etc. to cancer.
A point worth noting is that these inactive ingredients have both positive as well as negative effects on drugs. It is thus critical to focus on the positives and see how that aid in betterment can. This is exactly what the researchers aimed for and the results show it all. What has been observed in a majority of cases is that a drug isn’t able to do justice for what it is employed, says Daniel Reker, lead author of the research study and a former postdoc in the laboratory of Robert Langer. The reasons for this could be insufficient targeting, low bioavailability, or rapid drug metabolism, he added. Addressing this issue, MIT researchers aim to leave no stone unturned to get the drug in place and do a job that’s expected out of it.
How did the researchers work?
- The aim was clear – Develop a machine learning algorithm that’s capable enough of identifying self-assembling nanoparticles. For this, the prime requirement was to build a dataset for the algorithm.
Next up has to be the most critical aspect of the whole study. The researchers selected 16 self-aggregating small-molecule drugs. All of them had different chemical structures and applications. Additionally, they selected a set of about 90 widely available compounds. Ingredients right from additives that make the drug taste better to the ones that help in them
- last longer were included. Approval for the resulting nanoparticles wasn’t an issue because everything that went into the research process is already FDA-approved.
- Now, the team went on to test every combination of small-molecule drug and inactive ingredient.
With the screening platform showing all the desired results, talk about turning the machine learning platform on a much bigger library of compounds is doing rounds.
Going a step ahead, the researchers selected six nanoparticles for further validation. One among them is composed of sorafenib, a treatment that’s used for advanced liver and other cancers.
Though the prime aim of the screening platform is to optimize the efficiency of active drugs, the benefits are not limited to this. The platform holds the potential to customize inactive compounds as per the needs of the patients. Alternatives could also be provided.
The researchers are quite optimistic that the combination of machine learning and rapid screening will be able to predict interactions among different combinations of materials. Also, this could further accelerate the design of drugs and the nanoparticles to be able to deliver them throughout the body.