Conférence Iuliana IONITA-LAZA, Associate Professor Columbia University PDF Imprimer Envoyer
Mercredi, 10 Mai 2017 18:19
Conférence Iuliana IONITA-LAZA, Associate Professor Columbia University

FUN-LDA : A latent Dirichlet allocation model for predicting tissue-specific functional effects of noncoding genetic variation

Department of Biostatistics
Columbia University

Vendredi 12 mai 2017
12h45/13h45

Amphi 4 - Faculté de Médecine, 22 av. Camille Desmoulins - Brest

FUN-LDA: A latent Dirichlet allocation model for predicting tissue-specific functional effects

of noncoding genetic variation

Summary: Understanding the functional consequences of noncoding genetic variation is one of the most important problems in human genetics. We propose a latent Dirichlet allocation model for predicting functional effects of noncoding genetic variants (FUN-LDA) by integrating diverse epigenetic annotations for specific tissues and cell types from large scale genomics projects such as ENCODE and Roadmap Epigenomics. Our approach allows joint modelling of data from multiple tissues, and is easily extensible to data from additional tissues, not used to train the model. Using this unsupervised approach we predict tissue-specific functional effects for every position in the human genome. We demonstrate the usefulness of our predictions using several validation experiments. In particular, we provide a global view of the sharing of predicted functional variants across large number of tissues and cell types, and demonstrate that functional variants in promoters are more likely to be shared across many tissues compared with variants that fall in enhancers. Using eQTL data from the Genotype-Tissue Expression (GTEx) project we show that eQTLs in specific GTEx tissues tend to be most enriched among the predicted functional variants in relevant tissues in Roadmap. Furthermore, we show how these integrated functional scores can be used to derive the most likely causal tissue-/cell-type for a complex trait using summary statistics from genome-wide association studies. Finally, using experimentally validated functional variants from the literature, we show that our proposed method has better accuracy and precision in predicting functional variants compared to existing methods.

 
UBOCHU Brest CNRS Inserm EFS
Ifremer ENIB Telecom Bretagne