Trends & Insights: Researcher Highlights Summer Edition

DrugBank is passionate about equipping leading researchers with the data they need for new discoveries. As a trusted partner to numerous world-renowned researchers, we are proud to be contributing to groundbreaking research and aiding in solving some of the industry's most pressing issues. Our products help academic researchers drive innovation and push research further.

Our Curation team has put together a highlight of some impressive academic researchers, their work, and how they've powered their research with DrugBank's data.

Academic Research Highlight

Intense Bitterness of Molecules: Machine Learning for Expediting Drug Discovery

Eitan Margulis, Ayana Dagan-Wiener, Robert S.Ives, Sara Jaffari, Karsten Siems, Masha Y.Niv

Have you ever tried a new medicine and gone “yuck!” because it tasted so awful? Unfortunately, this is a common experience for many folks who take medication for the first time and end up having a not-so-good outlook on what should be making them feel better!

Formulating a drug product has many challenges. Flavour is one piece that isn't usually tested until further stages of clinical trials, meaning that if a new drug product ends up tasting foul, it will have to be reformulated, costing time and money. There are ingredients, like sweeteners, that can be used to hide the extreme bitterness, although they may not always be enough. Children are the biggest challenge when it comes to bitter-tasting medicines as they have a harder time ingesting medication that has an unpleasant flavour. An unintended consequence of poor-tasting medication can be the patient failing to complete the medical regimen and therefore receiving insufficient treatment. Geriatric patients also face the same difficulties and have been known to refuse to continue taking medications due to the unappealing flavour.

During clinical trials, compliance issues may arise from bitter-tasting drugs and reduce the number of viable results. This is where machine learning can come into play and help by detecting these bitter-tasting molecules earlier in the drug development process. Then, reformulation can occur as soon as the culprit compounds are determined and prevent any extra financial costs.

BitterIntense is a tool that can classify molecules into bitter and non-bitter with more than 80% accuracy on test datasets. It can also detect extremely bitter compounds, identifying around 25% of drugs falling into that category.

Making oral formulations is difficult, and by having this data available, it is easier for scientists to develop palatable products. The team that developed BitterIntense used DrugBank’s approved and experimental drug data to determine which drugs had very bitter compounds that would affect patient compliance. They found that 23.6% were very bitter from a list of 10,170 compounds. Of that, 18% were experimental drugs and 26% were approved drugs.

The group then looked into 34 potential drug treatments for COVID-19 to determine if bitter molecules were present. It was predicted that 41.2% had very bitter compounds, which offers insight that COVID-19 drugs are more bitter than a general list. However, having bitter compounds in COVID-19 treatments may not impair the medical regimen since patients will most likely have a loss of taste and smell.

Having this unique tool significantly helps with drug discovery as it allows for changes to the formulation to be made at earlier stages of the process. While patient compliance can be a tough challenge, there is hope that BitterIntense can reduce that barrier for all kinds of patients, whether they be kids or the geriatric population.

Read the paper

Academic User Highlight

Eitan Margulis

• Teaching assistant for Organic & General Chemistry
• Lab of Professor Masha Niv, part of the Hebrew University in Jerusalem.

Research Focus: Eitan's research focus falls within a few realms including biochemistry and food science, machine learning, Cheminformatics, Bioinformatics, and data science. He is currently focused on developing machine learning algorithms for taste prediction of molecules (drugs and food compounds included). He is working to predict the bitterness of drugs and is also interested in smell prediction. He aims to analyze and answer fundamental questions about taste perception by adapting ML models to chemical and biological information.

DrugBank Powered Research: DrugBank datasets act as an external dataset for models in the lab correlating taste with drugs. FooDB, which originated from the Wishart Lab, is also being used for this research.

Read the paper

Zombori Tamás

• Student at the University of Szegedlocated in Hungary

Research Focus: Zombori's interests are in Deep Learning and its intersection with drugs and healthcare. Within this focus Zombori and a team will be competing in the 2022 National NLP Clinical Challenges (n2c2) run by Harvard Medical School and George Mason University. In the competition, the team is given unstructured data about a patient’s drug usage. To solve the problem posed by the challenge, Zombori's team is building a named entity recognition (NER) with BERT.

DrugBank Powered Research: DrugBank’s data is being used to improve a deep learning model for the NER of drugs. He will use drug names and drug synonyms to strengthen the neural network.

Dr. Adam Amara

• Postdoctoral investigator and a computational and synthetic biologist at the Nutrition and Metabolism branch of the International Agency for Research on Cancer (IARC)

Research Focus: Dr. Amara is interested in science entrepreneurship and the intersection of computational biology, bioinformatics, and data science in the context of biotechnology and healthcare. He hopes to develop software solutions to tackle challenges in infection disease, microbiome, cancer, and biological engineering. As part of the IARC, he's investigating how cancer works in hopes of identifying potential cures for the disease, targets where action can be taken, and food associations.

DrugBank Powered Research: Dr. Amara hopes to use DrugBank for identifying potential products of degradation from drugs in our Metabolomics data, analyzing for correlations between drug-related product exposures and cancer risks, and to search for compounds outside our database.

Their goal is to publish a paper that details findings and cites DrugBank.

Researchers with a range of focuses from across the globe choose DrugBank as their source for reliable, verified data. From using our structured data to validate machine learning models to finding potential drug candidates, we offer data-driven solutions for many use cases.

DrugBank's database includes more than 21,000 drug-protein interactions, 130,000 drug product listings, drug-target data that includes proteins, gene identifiers, sequences, and so much more. With both paid and free academic licenses available, there are options to suit all of your research needs.

Learn more about our Academic datasets and see how we can help you get to your findings faster.