17-03-2014, 10:38 PM
Abstract
In this paper, we report systematic in depth analysis of
54 known pre-miRNA from Apis mellifera( Honey bee)
with a set of 14 attributes. We have derived this set of
attributes from secondary structure data that are
generated from pre-miRNA sequences from Apis
meillfera database using RNAfold. Principal
component analysis method has been applied for
dimension reduction. It reduces dimension of this set
from 54 to a set of 7. Out of these 7 only five
eigenvectors with variance more than 1.0 are
considered since other attributes showed a very low
variance. From this reduced set most dominating
attributed are identified using attributes ranking
computed by using weights of attributes and variance
of five eigenvectors. All attributes with rank more than
1.0 are selected. This cast the attributes set from
dimension 14 to 4 dominating attributes set. These
attributes can be used in pre-miRNA prediction model
that may facilitate miRNA biogenesis.
Keywords: miRNA, Secondary structure, PCA, premiRNA,
Variance
1. Introduction
miRNAs are an evolutionary conserved class of non
coding RNAs of small length approximately 20 to 25
nucleotides (nt) long and found in diverse organisms
like animal, plant etc. miRNAs play very important
role in various biological processes. They regulate
gene expression at post transcriptional level by
repressing or inactivating target genes [1, 2]. miRNA
biogenesis is highly associated with stem-loop feature
of its precursor’s secondary structure. As pre-miRNA
secondary structures consisting of stem-loop are highly
conserved across different species, extracting
informative attributes from secondary structure is
significant step in identification of miRNA from
unknown sequences [3, 4].
The biochemical based methodology used for
identification of novel miRNAs, in the laboratories can
be assisted by computational methods. Computational
methods can identify various potential miRNA that can
further be verified by the former approach. Therefore,
researchers have developed computational models for
miRNA prediction. Most of these methods form their
basis on set of features that can be used to identify
miRNA. Therefore, accuracy of prediction depends on
the dominance of the set of selected features and the
precision of the training methods. Different models
used in literature for prediction have been using
different set of features [5, 6, 7]. Further, the extraction
and selection of features also depends on the accuracy
of the secondary structure derived from the precursors.
There have been various techniques existing for
secondary structure prediction of RNA. These are also
applicable to derive structure of precursors. These
approaches are broadly classified into two categories –
energy based and grammar based. Grammar based
approaches originate from the field of linguistics. In
this approach rules that produce secondary structure
from a sequence are defined. Stochastic Context Free
Grammar (SCFG) has been used as a probabilistic
approach to model RNA secondary structure.
Contrafold is a
widely used web based SCFG implementation for
RNA secondary structure prediction [8].
The energy minimization techniques have been widely
used for RNA secondary structure prediction. These
techniques use the concept of dynamic programming
augmented with some standard rules of energy
computation. These rules compute the minimum
energy of a structure. MFOLD [9] and Vienna [10] are
energy minimization approach based packages
available in public domain.
In this paper, we report systematic in depth analysis of
54 known pre-miRNA from Apis mellifera( Honey bee)
with a set of 14 attributes. We have derived this set of
attributes from secondary structure data that are
generated from pre-miRNA sequences from Apis
meillfera database using RNAfold. Principal
component analysis method has been applied for
dimension reduction. It reduces dimension of this set
from 54 to a set of 7. Out of these 7 only five
eigenvectors with variance more than 1.0 are
considered since other attributes showed a very low
variance. From this reduced set most dominating
attributed are identified using attributes ranking
computed by using weights of attributes and variance
of five eigenvectors. All attributes with rank more than
1.0 are selected. This cast the attributes set from
dimension 14 to 4 dominating attributes set. These
attributes can be used in pre-miRNA prediction model
that may facilitate miRNA biogenesis.
Keywords: miRNA, Secondary structure, PCA, premiRNA,
Variance
1. Introduction
miRNAs are an evolutionary conserved class of non
coding RNAs of small length approximately 20 to 25
nucleotides (nt) long and found in diverse organisms
like animal, plant etc. miRNAs play very important
role in various biological processes. They regulate
gene expression at post transcriptional level by
repressing or inactivating target genes [1, 2]. miRNA
biogenesis is highly associated with stem-loop feature
of its precursor’s secondary structure. As pre-miRNA
secondary structures consisting of stem-loop are highly
conserved across different species, extracting
informative attributes from secondary structure is
significant step in identification of miRNA from
unknown sequences [3, 4].
The biochemical based methodology used for
identification of novel miRNAs, in the laboratories can
be assisted by computational methods. Computational
methods can identify various potential miRNA that can
further be verified by the former approach. Therefore,
researchers have developed computational models for
miRNA prediction. Most of these methods form their
basis on set of features that can be used to identify
miRNA. Therefore, accuracy of prediction depends on
the dominance of the set of selected features and the
precision of the training methods. Different models
used in literature for prediction have been using
different set of features [5, 6, 7]. Further, the extraction
and selection of features also depends on the accuracy
of the secondary structure derived from the precursors.
There have been various techniques existing for
secondary structure prediction of RNA. These are also
applicable to derive structure of precursors. These
approaches are broadly classified into two categories –
energy based and grammar based. Grammar based
approaches originate from the field of linguistics. In
this approach rules that produce secondary structure
from a sequence are defined. Stochastic Context Free
Grammar (SCFG) has been used as a probabilistic
approach to model RNA secondary structure.
Contrafold is a
widely used web based SCFG implementation for
RNA secondary structure prediction [8].
The energy minimization techniques have been widely
used for RNA secondary structure prediction. These
techniques use the concept of dynamic programming
augmented with some standard rules of energy
computation. These rules compute the minimum
energy of a structure. MFOLD [9] and Vienna [10] are
energy minimization approach based packages
available in public domain.