Asier Larrea (Spain)

Biofisika Institute Biochemistry and Molecular Biology
My scientific formation began in 2014, when I started my studies in the Bachlelor´s degree in Biotechnology at the University of the Basque Country. In 2016 I joined the research group of Doctor Cesar Martin in the Department of Biochemistry and Molecular Biology of the UPV/EHU and the Biofisika Institute, where I carried out the study of the mechanisms involved in the development of Familial Hypercholesterolemia until I graduated with Excellence. In 2019 I studied the Master in Molecular Biology and Biomedicine (UPV/EHU-UC) and in 2020 I started my PhD studies with a Basque Government scholarship. In this period I have obtained many research grants (IkasIker, JAE Intro) and I have published 7 cientific articles.

Author Of 3 Presentations

Live Q&A (ID 1550)

O005 - MLb-LDLr: A Machine Learning Model for Predicting the Pathogenicity of LDL receptor Missense Variants (ID 471)

Session Type
Late Breaking Sessions
Session Time
11:00 - 12:30
Date
Mon, 31.05.2021
Room
Hall A (Live Q&A)
Lecture Time
11:28 - 11:35

Abstract

Background and Aims

Cardiovascular disease (CVD) is the leading cause of death worldwide and is often related to high plasma concentrations of low-density lipoprotein cholesterol (LDL-c). One of the most frequent dyslipidaemias is familial hypercholesterolemia (FH), which is mostly caused by mutations in the LDL receptor (LDLr). Although an early identification is essential to reduce premature mortality, only the 10 % of FH patients are properly diagnosed.

In vitro characterization of variants is time consuming and expensive, so computational predictors of pathogenicity of mutations are under constant development. The aim of our work was to create a machine learning-based model that can predict the pathogenicity of LDLr missense variants, the most common ones, on an easy and reliable way.

Methods

Using more than 700 LDLr missense variants characterized on ClinVar database and Excel solver Evolutionary algorithm, we created a Machine Learning model that predicts the pathogenicity of a variant based on seven characteristics of the mutated amino acid: Conservation of the residue, original and substituting amino acid, hydrophobicity, size, charge and affected domain.

We used a part of the dataset as training group for the obtention of pathogenicity frequency distribution and other essential parameters. Then, the model was tested with the other part of the dataset, the validation group.

Results

MLb-LDLr shows a sensitivity of 92.5% and a specificity of 91.5%, matching or even surpassing other predictor software such as PolyPhen-2 or SIFT.

Conclusions

With an accuracy higher than 90%, we conclude that in silico predictions are a reliable source of information about LDLr variant pathogenicity.

Hide

O057 - Lipoprotein(a) in heterozygous familial hypercholesterolemias. Influence of the causative gene and type of mutation (ID 475)

Session Type
Genetics
Session Time
16:00 - 17:30
Date
Tue, 01.06.2021
Room
Live Streamed
Lecture Time
17:05 - 17:13

Abstract

Background and Aims

Background:Lipoprotein(a) [Lp(a)] is a recognized cardiovascular risk factor. Lp(a) concentration in heterozygous familial hypercholesterolemia (heFH) is not well established. Whether the genetic defect responsible for heFH plays a role in determining Lp(a) concentration is unknown.

Aims:To study Lp(a) concentration in subjects genetically diagnosed with heFH and to assess the influence of the genetic defect responsible for heFH on its concentration.

Methods

Methods: Cross-sectional study, performed in a lipid clinic in Spain. We studied 511 heFH adults according to the responsible gene (LDLR, APOB, APOE and PCSK9). We selected 443 subjects LDLR, 27 subjects APOB, 37 subjects carriers of the p.(Leu167del) mutation in APOE, and 4 subjects PCSK9.

Results

Results:

Lipid levels differed across gene groups after adjusting for age, sex, and BMI.Lp(a) concentration differed among subjects with LDLR, APOB, and APOE mutation(p <0.001).Median Lp(a) concentration was greatest in APOB-dependent FH 36.5 mg/dL(IQR 22.0,60.8),intermediate in LDLR-dependent FH,21.7 mg/dL (IQR 17.9, 26.4)(and independent on the affected LDL receptor protein domain) and lowest in carriers of the p.(Leu167del) mutation in APOE, 7.9 mg/dL (IQR 4.9,12.7).Lp(a) geometric means, adjusted for age, sex, and BMI differed significantly. The geometric mean of LPA KIV-2 repeats did not differ among the FH gene subgroups and the estimations and differences for Lp(a) remained unchanged after adjustment for the number of KIV-2 repeats.

Conclusions

Conclusions: The concentration of Lp(a) in heFH is depending on the responsible gene. Lp(a) concentration was gratest in APOB- dependent FH. In LDLR-dependent, FH Lp(a) levels are not different depending on the affected protein domain.

Hide

Presenter Of 2 Presentations

Live Q&A (ID 1550)

O005 - MLb-LDLr: A Machine Learning Model for Predicting the Pathogenicity of LDL receptor Missense Variants (ID 471)

Session Type
Late Breaking Sessions
Session Time
11:00 - 12:30
Date
Mon, 31.05.2021
Room
Hall A (Live Q&A)
Lecture Time
11:28 - 11:35

Abstract

Background and Aims

Cardiovascular disease (CVD) is the leading cause of death worldwide and is often related to high plasma concentrations of low-density lipoprotein cholesterol (LDL-c). One of the most frequent dyslipidaemias is familial hypercholesterolemia (FH), which is mostly caused by mutations in the LDL receptor (LDLr). Although an early identification is essential to reduce premature mortality, only the 10 % of FH patients are properly diagnosed.

In vitro characterization of variants is time consuming and expensive, so computational predictors of pathogenicity of mutations are under constant development. The aim of our work was to create a machine learning-based model that can predict the pathogenicity of LDLr missense variants, the most common ones, on an easy and reliable way.

Methods

Using more than 700 LDLr missense variants characterized on ClinVar database and Excel solver Evolutionary algorithm, we created a Machine Learning model that predicts the pathogenicity of a variant based on seven characteristics of the mutated amino acid: Conservation of the residue, original and substituting amino acid, hydrophobicity, size, charge and affected domain.

We used a part of the dataset as training group for the obtention of pathogenicity frequency distribution and other essential parameters. Then, the model was tested with the other part of the dataset, the validation group.

Results

MLb-LDLr shows a sensitivity of 92.5% and a specificity of 91.5%, matching or even surpassing other predictor software such as PolyPhen-2 or SIFT.

Conclusions

With an accuracy higher than 90%, we conclude that in silico predictions are a reliable source of information about LDLr variant pathogenicity.

Hide