| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 984.78 KB | Adobe PDF |
Advisor(s)
Abstract(s)
A complexidade genética e clínica que caracterizam a per turbação do espetro
do autismo (PEA) têm limitado o desenvolvimento de biomarcadores que
permitam um diagnóstico precoce e um prognóstico fiável, assim como uma
abordagem personalizada para a inter venção terapêutica. Neste estudo
pretendeu-se desenvolver uma abordagem integrativa para predição da
apresentação clínica baseada em informação de variantes genéticas (Copy
Number Variants, CNVs), com aplicação clínica no diagnóstico e prognóstico
na PEA. Para tal, técnicas de aprendizagem automática (machine learning)
foram aplicadas a dados clínicos e genéticos de 2446 doentes com PEA,
recrutados no âmbito do consórcio Autism Genome Project. Análise de
clustering de dados clínicos multidimensionais definiu, nesta população,
dois subgrupos de pacientes com per fis clínicos diferindo significativamente
em termos de capacidade verbal, nível cognitivo, gravidade da doença e
compor tamento adaptativo. A análise dos CNVs que afetam especificamente
genes do cérebro, nos mesmos indivíduos, identificou 15 processos biológicos
enriquecidos em genes alterados. A aplicação de um algoritmo de
machine learning para classificação dos doentes com apresentação clínica
mais disfuncional, com base nos processos biológicos alterados, mostrou
que correlações entre fenótipo clínico e biologia subjacente são possíveis
na PEA e que, para grupos populacionais com dados informativos, existe
um poder preditivo razoável. Para implementação deste conceito na prática
clínica serão necessários estudos mais alargados com dados clínicos e
genómicos mais completos.
The genetic and clinical complexity that characterize Autism Spectrum Disorder (ASD) has hindered the development of biomarkers for early diagnosis and reliable prognosis, as well as a personalized to therapeutic inter vention. This study aimed to develop an integrative approach for clinical presentation prediction based on Copy Number Variants (CNVs), with clinical application for diagnosis and prognosis of ASD. For this purpose, machine learning techniques were applied to a dataset of 2446 patients with ASD, recruited by the Autism Genome Project. Clustering analysis of multidimensional clinical data allowed the definition of two patient subgroups in this population, with clinical profiles dif fering significantly in verbal ability, cognitive level, disease severity and adaptive behavior. In the same subjects, analysis of CNVs specifically af fecting brain-expressed genes identified 15 biological processes enriched for the disrupted genes. A machine learning algorithm was trained and tested to classif y patients with more dysfunctional clinical presentation based on altered biological processes. The results showed that correlations between clinical phenotype and underlying biology can be established in ASD and that, for datasets with suf ficiently informative data, there is a reasonable predictive power. Fur ther studies with more complete clinical and genomic data are needed to implement this concept in clinical practice.
The genetic and clinical complexity that characterize Autism Spectrum Disorder (ASD) has hindered the development of biomarkers for early diagnosis and reliable prognosis, as well as a personalized to therapeutic inter vention. This study aimed to develop an integrative approach for clinical presentation prediction based on Copy Number Variants (CNVs), with clinical application for diagnosis and prognosis of ASD. For this purpose, machine learning techniques were applied to a dataset of 2446 patients with ASD, recruited by the Autism Genome Project. Clustering analysis of multidimensional clinical data allowed the definition of two patient subgroups in this population, with clinical profiles dif fering significantly in verbal ability, cognitive level, disease severity and adaptive behavior. In the same subjects, analysis of CNVs specifically af fecting brain-expressed genes identified 15 biological processes enriched for the disrupted genes. A machine learning algorithm was trained and tested to classif y patients with more dysfunctional clinical presentation based on altered biological processes. The results showed that correlations between clinical phenotype and underlying biology can be established in ASD and that, for datasets with suf ficiently informative data, there is a reasonable predictive power. Fur ther studies with more complete clinical and genomic data are needed to implement this concept in clinical practice.
Description
Keywords
Perturbação do Espetro do Autismo Variantes Genéticas Perturbações do Desenvolvimento Infantil e Saúde Machine Learning Copy Number Variants (CNVs)
Pedagogical Context
Citation
Boletim Epidemiológico Observações. 2018 maio-agosto;7(22):38-42
Publisher
Instituto Nacional de Saúde Doutor Ricardo Jorge, IP
