Have you ever wondered whether the drugs you consume daily, such as blood pressure medications and antihistamines, interact with one another? Around 8.6 million people in Europe are admitted to hospitals yearly due to adverse reactions to drugs, five per cent of which are caused from drug-to-drug interactions. Although drugs are tested alongside other medications, more than 50 per cent of adverse drug reactions are detected after the drug is released into the market.

This research is investigating the use of knowledge graphs and how these can be used to capture more complex relationships between drugs based on several properties to predict potentially unknown drug-drug interactions. Data is obtained from several well-known repositories and the information is represented as a heterogeneous structure.

In this knowledge graph, biomedical concepts such as drugs, diseases, and side effects are represented as entities, while the edges in the graph denote their respective relationships. For example, an edge between two drug entities might indicate a potential drug-to-drug interaction, while an edge between a drug and a disease entity might describe a treatment.

A graph auto-encoder model was trained to predict whether a potential adverse interaction exists among two given drugs. The first phase of the auto-encoder, which is known as the encoder, was implemented using graph neural networks  (GNN) and embeds the drug data from the knowledge graph into a low-dimensional space, representing each drug as a vector. The decoder then uses the generated drug embeddings to train a deep neural network that learns to distinguish between interacting and non-interacting drug pairs.

GNN, a relatively new area in the field of deep learning, are a type of neural network that can extract meaningful information from graphs and then perform inference on data represented by graphs.

The resulting model obtained a precision of 91.95 per cent on the evaluation dataset. Knowledge graphs could provide valuable insights from the data on why drugs may interact with one another.

Moreover, GNN can generate better drug representations than other models due to their ability to consider all types of relations between the entity links, path, and substructure information.

Further research could potentially recommend other drugs used for the same treatment, which do not pose the same interaction risk and thus help doctors provide better and safer healthcare.

Lizzy Farrugia is a Master in Artificial Intelligence student at the University of Malta. The research is being led by Charlie Abela and Jeremy Debattista, from the Department of Artificial Intelligence within the Faculty of ICT.

Sound Bites

•        Researchers have discovered a unique class of medications that can be used to treat patients with a variety of coronary issues, including heart attacks. Ticks are used as a model for developing these drugs that prevent blood clotting, as they have naturally occurring blood clot inhibitors. Their saliva contains pharmacological and immunological active compounds, creating an immune response in the body and act as blood thinners. 

•        A new study of brain scans of boys and girls with autism found significant differences in brain centres, suggesting that gender-specific diagnostics are needed. Historically, research was biased towards male patients, with diagnosis being four times more common in boys than girls. These differences might explain why girls with autism have fewer overt repetitive behaviours than boys. It will help girls with autism get diagnosed earlier, since treatments work best during the preschool years when the brain’s motor and language centres are developing.

For more sound bites listen to Radio Mocha www.fb.com/RadioMochaMalta/.

DID YOU KNOW?

•        Drug, supplement or foods affect how medication stays in the body, often by stimulating or inhibiting the production of specific enzymes in the liver or intestine.

•        The enzymes play an important role in metabolising drugs and interactions can make a drug more powerful, whereby a standard dose can become an overdose or it can be rendered less potent or ineffective.

•        Several foods can also block or stimulate the enzymes that break down drugs.

•        Large amounts of grapefruit juice inhibit the enzyme that clears statins, the main ingredient in cholesterol-lowering medication.

•        The human body also breaks down drugs differently. Increased age, liver or kidney disease will slow down the metabolisation of drugs.

For more trivia, see: www.um.edu.mt/think.