Classification of Promoters based onChromatin Features

information crucial for the proper development and functions of an organism. But there
also exist mechanisms, external to the underlying DNA sequence of an organism that take
part in regulating gene expressions in different cell types. These external mechanisms are
typically called the epigenetic factors. Their examples include changes in the chromatin
structure through histone modifications, where histones are the chief protein components
of chromatin.
Study of epigenetic factors has gained interest and impetus in the last couple of
years due to their roles in diseases. Their precise role in gene regulation is still poorly
understood. Genome-wide maps of various histone modifications across various cell-types
are now publicly available. In this work we have focussed on high-resolution maps of 21
histone methylations from Barski et al. [1] in CD4+ T cells. In this talk, we will discuss
the power of these histone modifications in predicting gene expression levels.
Using support vector machines (SVMs), we show that histone modifications at
promoters achieve near perfect accuracy in predicting the gene expression levels in CD4+
T cells. The results by SVM are in congruence with experimental evidences in biology.
Now such high resolution histone maps are impractical to achieve in the hundreds of
different mammalian cell-types. We therefore looked at the generality of the classifier in
predicting gene expressions in other cell-types, but using the same CD4+ histone marks.
Surprisingly, we get a remarkable accuracy when predicting expressions in other tissues
as well.
Independently, we have developed a multi-class classifier that uses Markov modelbased
sequence-features from promoters to predict the tissue-specific expression of their
downstream genes. We propose combining the two classifiers in an enhanced SVM, which
we believe will be even more powerful in predicting tissue-specific expression patterns of
genes.
References