Discovering Motifs in "Omics" Signals using Local Clustering of Curves

The Institute of Economics will hold the next meeting of its Seminar Series on Tuesday, March 14, 2016: Marzia Cremona, from The Pennsylvania State University, will present the paper Discovering Motifs in "Omics" Signals using Local Clustering of Curves.

Abstract:

Functional Data Analysis (FDA) can be broadly employed to exploit the heterogeneous, high-dimensional and complex “Omics” data generated by Next Generation Sequencing technologies. The core idea consists in considering “Omics” data at high resolution, treating them as “curves” of measurements along the DNA sequence.

In this framework we develop probabilistic K-mean with local alignment, an algorithm for clustering a set of curves based on similar curve pieces they may share. This novel methodology brings together ideas from FDA (jointly cluster and align curves), bioinformatics (form local alignments expanding high similarity “seeds”) and fuzzy clustering (each curve can belong to more than one cluster, i.e. contain more than one “typical” curve piece).

We employ the algorithm to discover functional motifs in “Omics” signals related to mutagenesis and genome dynamics, exploring high-resolution profiles of different mutation rates in regions of the human genome where these rates are globally elevated.