How AI is helping to predict disease-related mutations
Artificial intelligence (AI) has started to play a vital role in uncovering secrets behind biological data, with a University of Hong Kong research team becoming the first to use a deep-learning approach to predict disease-associated mutations of metal-binding sites in body proteins. Such mutations are associated with a range of human health problems, depending on the metal, including prostate, muscular, metabolic, and immune system diseases.
The research was led by Professor Hongzhe Sun (Croucher Senior Research Fellowship 2010) from the Department of Chemistry, in collaboration with Professor Junwen Wang from the Mayo Clinic, Arizona.
The study discovered that mutations in the human genome are strongly associated with different diseases. If such changes occur in the coding region of DNA, this may disrupt the metal-binding sites of proteins.
Physiological metals, such as zinc, iron and copper, play pivotal roles in our lives, and their concentration in cells must be strictly regulated. An excess or deficiency can cause severe diseases in humans.
The findings were recently published in top scientific journal Nature Machine Intelligence.
The research team first integrated “omics” data – approaches spanning biological fields that end in “omics” for genetic and molecular profile analysis – from different databases to build a comprehensive training dataset.
Following this, statistics from the collected data showed the team that different metals had different disease associations.
Mutations in calcium and magnesium-binding sites are associated with muscular and immune system diseases, respectively, while zinc has a major role in breast, liver, kidney, immune system and prostate diseases. Iron-binding site mutations are more associated with metabolic diseases. Mutations of manganese and copper-binding sites are associated with cardiovascular diseases. The latter is also associated with nervous system diseases.
The researchers then used the novel method of extracting spatial features from metal-binding sites using an energy-based affinity grid map. These features were merged with physicochemical sequential features to train the model, with final results indicating enhanced performance and accuracy of predictions.
The advanced deep-learning approach offers a new way to integrate experimental data with bioinformatics analysis, with the potential for the technology to help scientists predict DNA mutations associated with cancer, cardiovascular diseases, genetic disorders, and other diseases.
Understanding of disease-associated mutations at metal-binding sites of proteins can also facilitate the discovery of new drugs.
Professor Sun Hongzhe is a leading expert in the field of biological inorganic chemistry and metallomics. Major achievements include studies of metal transport and storage proteins and their relationship with metallodrugs, providing a basis for mechanism-based drug design. Sun received his PhD from the University of London (Birkbeck) in 1996. After two years as a Research Fellow at the University of Edinburgh, he joined the Department of Chemistry at the University of Hong Kong as an Assistant Professor. He was promoted to Professor in 2007. He received his Croucher Senior Research Fellowship in 2010.
To view Professor Sun Hongzhe’s Croucher profile, please click here.