AI ​​Tool Can Predict How Drugs Will React In the Human Body

Original article was published by Mantha Anirudh on Artificial Intelligence on Medium

AI ​​Tool Can Predict How Drugs Will React In the Human Body

AI ​​Tool Can Predict How Drugs Will React In the Human Body

It’s a wonder that a new in-depth learning-based AI tool named ‘Metabolic Translator’ could soon provide researchers with a good handle on how drugs in development in the human body will work. Drug companies go through extensive tests to see what you can do.

The metabolic translator can predict metabolites and helps in improving the process of interaction products between small molecules such as enzymes and drugs.

The new AI tool will take advantage of massive reaction datasets and deep-learning methods to provide developers with a broader picture of what a drug can do in the human body. This method is not regulated by rules that companies use to determine metabolic reactions, paving the way for new innovations.

Lydia Kavraki, a professor of electrical and computer engineering, bioengineering, computer science, and mechanical engineering at Rice’s Ken Kennedy Institute, said that “When you are striving to confirm that a compound is a potential drug, you should check for toxicity”. She also says that “You want to confirm what this is going to do, but you still need to know what’s going on”.

Researchers have trained the metabolite translator to predict metabolism by any enzyme, but measure its success against current regulatory-based methods that focus on enzymes in the liver. These enzymes are highly responsible to detox and remove xenobiotics such as pollutants, pesticides, and drugs. However, metabolism is also caused by other enzymes.

A graduate student and lead author named ‘Ellen Litsa’ said that “Human bodies are networks of chemical reactions,”. They have enzymes that work on chemicals and can break down or form bonds that make their structures toxic or cause other problems. Existing methods concentrate on the liver as most xenobiotic compounds are metabolized there. We are trying to capture human metabolism in general, without any work.

“The safety of a drug depends not only on the drug but also on the metabolism that takes place when the drug is processed in the body,” says Litsa. She also says the growth of machine learning structures that work on structural data like chemical molecules makes work possible.

In 2017, Transformer was launched as a sequence translation method that has determined widespread use in language translation and is depended on SMILES (Simplified molecular-input line-entry system”), a code that uses plain text instead of diagrams to represent chemical molecules.

Litsa says, “What we are doing is equivalent to translating a language like English from German”. In the absence of experimental data, the laboratory used transfer practice to develop the metabolite translator. They first trained 900,000 known chemical reactions on the transformer model and then fine-tuned them with human metabolic transformation data.

Metabolite translator results were compared by researchers with several other assessment methods by analyzing known SMILES sequences of 179 metabolic enzymes and 65 drug drugs.

Although they trained for metabolite translation on a simple dataset specific to met drugs, it demonstrated protocol-based methods developed specifically for drugs. But it also detects enzymes that are not normally involved in methadone metabolism and are not detected by existing methods.

We have a system that can be assessed on a par with rule-based systems, and we do not place in our system any rules that require manual work and expert knowledge”. Using an ML-based approach, we train a system to understand human metabolism without the requirement to encode this knowledge in terms of rules explicitly. But, two years ago this work was not possible”, Kavraki said.

If you would like to integrate AI tools in your business, please contact us.