Solving the secret of a healthy mouth

14 October 2016 – Healthy mouths differ greatly from each other in their saliva and microbiota composition. Applying machine-learning algorithms to complex –omics data can help people implement personalized prevention approaches to oral disorders into their daily dental hygiene. This is the main conclusion of research by TiFN PhD fellow Sultan Imangaliyev. The computational biologist defended his thesis on October 14, at the University of Amsterdam. His work has provided leads for the development of next-generation strategies to maintain and improve oral health. 

In Europe, dental care is one of society’s three highest medical costs. Although people visit the dentist, on average, once or twice a year, oral diseases remain one the most common diseases worldwide, with a large number of adults and the elderly suffering from periodontitis, gingivitis and caries. It is unclear why some people have a ‘healthy mouth’, and others – despite proper hygiene – still have oral health issues.

Machine learning
TiFN’s Oral Health research programme focusses on the biological interactions between saliva, oral microbiota and the human defence system – studies that provide huge amounts of complex data which cannot be analysed via conventional statistical methods such as Principal Component Analysis. Instead, Imangaliyev creatively applied a range of modern machine learning approaches, including single-view machine learning, multi-view machine learning and deep learning. Multi-view machine learning techniques, unlike single-view, take advantage of multiple data sources, integrating them in a single model, so maximizing the overall model performance. Deep-learning techniques extract the most useful variable representations via series of nonlinear transformations computed in multiple layers of deep, artificial neural networks. “Although machine learning and descriptive statistics both recognize patterns in large datasets, machine learning is able to use these recurring patterns to predict future data. This helps to save money and time by, for example, avoiding treatments which might not work”, Imangaliyev explains.

First steps towards understanding
“Machine learning has helped us to take a first step towards understanding what makes a mouth healthy. From an ecosystemic point of view, a healthy mouth is a system where all the components are in dynamic homeostasis”, says the scientist. “A disease is a result of a dysbiotic ecological shift caused not by a single microorganism, but by a network of interacting species.” Via application of machine learning to metabolomics, metagenomics and biochemistry data, Imangaliyev and his colleagues identified subgroups of healthy individuals with different compositions of oral microbiota and salivary biochemicals. “We also carried out a clinical trial in which people did not brush their teeth for two weeks. After only two days, some of them had significant amounts of plaque whereas others still had smooth, clean teeth”, says the scientist.

Targeted approach
Follow-up research is needed but, according to Imangaliyev, there is clearly no magic bullet for improving oral health: “Dentistry needs to adopt a more targeted, personalized approach in the prevention of, for example, gingivitis.” Imangaliyev is content with the research and its outcomes. “Applying machine learning on such a large panel of heterogeneous datasets is unique in oral health research, as is TiFN’s approach of focussing on healthy rather than diseased people”, he says.

The computational biologist, who will continue his career as a post-doc scientist at the VU Medical Centre, experienced his time at TiFN as an exciting rollercoaster-ride. “I learned a lot about biological systems, and improved my personal skills”, he says. “Biological systems do not come with a user manual. You have to find out yourself how they work and then fix them yourself. This requires courage and perseverance.”