The Inverse Problems, Optimisation and Machine Learning Group at the University of Oviedo, led by Professor Juan Luis Fernández-Martínez, in collaboration with Professor Andrzej Kloczkowski of the Batelle Centre of Mathematical Medicine, under the Columbus (Ohio) Nationwide Children Hospital, has developed an algorithm that enables scientists to determine whether a mutation in a protein's amino acid chain can destabilise it and cause disease, surpassing the reliability of existing methods. Knowledge about how stable or unstable this change is (called polymorphism) is a key aspect in drug design and in the study of numerous diseases. This work is part of Óscar Álvarez's doctoral thesis, which is undertaken in the Department of Physical and Analytical Chemistry at the University of Oviedo. The results have come to light in the "Biomolecules" journal. 

 

Proteins are molecules formed by chains of amino acids and are essential for the functioning of the organism of living beings. With a small number of amino acids (only 20 different), millions of proteins are generated, and their sequence determines their structure and functionalities. Proteins are synthesised depending on how the genes that encode them are regulated. 

 

Understanding how one or several mutations simultaneously affect the functionalities of a protein is one of the ongoing problems in genomics, since a single substitution in the amino acid chain can cause adverse effects. The development of many diseases depends on whether these mutations destabilise proteins, changing their spatial structure. Predicting such a change is therefore a fundamental aspect in the study of diseases and in the search for drugs that can prevent this effect.

 

The study of diseases requires knowledge about the genome and the possible gene mutations in an individual as well as how they impact the stability of the proteins that these genes encode. Machine learning techniques have therefore been used by consensus. "It is assumed that different mutations have a characteristic curve of energy variations. These curves are learned from different experimental databases and are subsequently used to predict the effect of new mutations in one or several positions. As the database of the effect of mutations is improved, the reliability of the designed method increases," says Professor Fernández-Martínez.  

 

The same group developed a model to explain how the main mutations in patients with chronic lymphocytic leukaemia impacted the transcriptome (the set of RNA molecules present in a cell), affecting genes that regulate the immune system. Professor Fernández-Martínez and Óscar Álvarez explain that the techniques used in this research "are based on the use of consensus; that is, different machine learning models are trained in parallel so that, by validating the method with independent data, the degree of destabilisation is obtained and uncertainty can be quantified using a decision algorithm by majority vote. These methods are used in the optimal design of drugs that are coupled to the protein by inhibiting it."

 

This project began in 2013 during a visit by Professor Fernández-Martínez to the Battelle Centre for Mathematical Medicine Modelling at the Ohio hospital, in which the University of Oviedo Inverse Problems Group actively collaborates. For the researcher, "the structure of bio-sanitary research in hospitals in the United States is a model: engineers, physicists, mathematicians, biologists, biochemists and doctors working together against diseases at the translational level; that is, producing results on the computer and in the laboratory and taking them to the hospital. We are light years away, but this organisation will also be adopted here, or the health system will be outdated," he concludes.

 

Reference