Cardiovascular diseases are the leading cause of death worldwide. In some cases, these diseases may be associated with a poor diet or a sedentary lifestyle, but in other cases it is genetic causes that are responsible for these diseases. For example, Familial Hypercholesterolaemia (FH) is mainly due to mutations occurring in certain genes, and the associated high cholesterol levels significantly increase the risk of developing cardiovascular disease.

Most cases of familial hypercholesterolaemia are due to the LDLr gene. More than 3,000 variants of the gene have already been described. The LDLr protein is a protein present in cell membranes that undertakes to internalise cholesterol; if there is a problem in this protein, the cholesterol in the blood cannot be internalised, so it accumulates in the blood and leads to a range of diseases.

So “this work has set out to develop software to specifically analyse the LDLr gene, the main cause of Familial Hypercholesterolaemia, in order to assist in the early diagnosis of the disease and to find a specific treatment", said Asier Larrea-Sebal, a researcher in the UPV/EHU's Molecular Mechanisms of Familial Hypercholesterolemia research group and lead author of the work.

To do this, "we studied many of the LDLr gene variants that have already been collected and characterised in the ClinVar database, and developed an advanced machine learning algorithm to accurately predict the pathogenicity of the LDLr variants. To do this, various characteristics of the protein were taken into account: position of the mutation, size, etc.", explained the UPV/EHU researcher.

Asier Larrea-Sebal pointed out that "a major challenge in the optimization process of the MLb-LDLr software was to create a balanced software program capable of accurately predicting pathogenic and benign variables. The strength of the software developed in this work lies in its accuracy. In fact, by taking into account both benign and pathogenic mutations, it makes predictions with an accuracy in excess of 90 %. "Right now, to predict the impact of a given mutation on protein activity, software capable of predicting both pathogenic and benign mutations is essential. In fact, the instruments with the highest precision in pathogenic variants have low precision in benign variables," explained Asier Larrea-Sebal.

In addition, "the MLb-LDLr software allows the database to be updated and therefore new variants to be incorporated into the ClinVar database. That enables the accuracy of the database to be continuously increased in order to make an updated forecast for each new variant described", said the UPV/EHU researcher.