Luogo: Aula Seminari "-1" - Dipartimento di Matematica - Via Sommarive 14 - Povo - Trento
- Giulia Tini - PhD student in mathematics
Technological advances in measuring and collecting biological data, such as genes and proteins (i.e. omics) allow researchers to investigate those biological layers together for their association with a phenotypic trait of interest.
In some cases, interactions occurring between different omics are known: we performed integration based on prior knowledge to combine gene expression and methylation levels from adipocytes, which revealed the role of methylation in regulating genes during adipogenesis.
However, interactions among omics data are not always known, suggesting the use of unsupervised statistical methods. We studied the impact of method choice, feature selection, number of integrated data types on integration results by comparing several statistical methods.
Finally, we propose a combined framework to improve multi-omics integration performance by taking advantage of the information provided by prior knowledge and of the statistical methods ability to search for unknown interactions.
Supervisor: Corrado Priami