Subsampling methods for genomic inference

Large-scale statistical analysis of data sets associated with\ngenome sequences plays an important role in modern biology. A\nkey component of such statistical analyses is the computation of\np-values and confidence bounds for statistics\ndefined on the genome. Currently such computation is commonly\nachieved through ad hoc simulation measures. The method of\nrandomization, which is at the heart of these simulation\nprocedures, can significantly affect the resulting statistical\nconclusions. Most simulation schemes introduce a variety of\nhidden assumptions regarding the nature of the randomness in the\ndata, resulting in a failure to capture biologically meaningful\nrelationships. To address the need for a method of assessing the\nsignificance of observations within large scale genomic studies,\nwhere there often exists a complex dependency structure between\nobservations, we propose a unified solution built upon a data\nsubsampling approach. We propose a piecewise stationary model\nfor genome sequences and show that the subsampling approach\ngives correct answers under this model. We illustrate the method\non three simulation studies and two real data examples.

1. 1
   An integrated encyclopedia of DNA elements in the human genome

   Nature201219,311 citationsDataRank 23.8Top 3%
2. 2
   Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project

   Nature20075,248 citationsDataRank 15.6Top 12%
3. 3
   Unsupervised pattern discovery in human chromatin structure through genomic segmentation

   Nature Methods2012651 citationsDataRank 0.972
4. 4
   Classification of human genomic regions based on experimentally determined binding sites of more than 100 transcription-related factors

   Genome Biology2012267 citationsDataRank 9.0Top 23%
5. 5
   The role of chromatin accessibility in directing the widespread, overlapping patterns of Drosophila transcription factor binding

   Genome Biology2011240 citationsDataRank 0.823