Applications & Challenges of Deep Learning in the field of
bioinformatics
Pandya Mrudang Daxeshkumar
Assistant professor, Information Technology dept.
CHARUSAT
Anand,India
E-mail- mrudangpandya.it@charusat.ac.in
Dr. Jardosh Sunil
Principal Software Engineer
Progress Software,
Hyderabad, India
E-mail: sunil.jardosh@gmail.com
Abstract: Analyzing large (big) data is becomes one of
the most emerging research areas for decision making.
Deep Learning (DL), a recent development in machine
learning techniques has demonstrates promising results
in many fields like computer vision, speech recognition
and bioinformatics. In the field of bioinformatics, the
transformation of medical domain big data into
measurable knowledge becomes one of the most
challenging tasks. In our work we have analyzed
applicability of DL to the bioinformatics domain,
current state-of-art and challenges. We believe our
comprehensive analysis provides profound insights on
DL architectures and methods used in bioinformatics.
Keywords: Machine Learning, Deep Learning,
Bioinformatics, Deep Neural Network (DNN),
Multilayer Perceptron (MLP), Deep Belief Network
(DBN), Stacked Auto-Encoder (SAE), Convolution
neural network (CNN), Convolutional Autoencoder
(CAE), Deep Spatiotemporal Neural Networks (DST-
NN).
1. Introduction
Machine-Learning becomes vitally important for
decision making systems, recommendation systems,
image analysis and web searching. There are several of
Machine Learning Algorithms(i.e., Artificial neural
network, support vector machines, random forests, hidden
Markov models, Bayesian networks, Gaussian networks)
have been applied to various bioinformatics fields like
genomics, proteomics and systems biology [13].
Usefulness of various ML techniques in various
bioinformatics fields are observable by resent publication
trend shown in figure1. It shows that in last few years
use of deep learning techniques in bioinformatics domain
growing very fast and it is quite evident in some of the
recent systems IBM’s watson[3] and Google’s
Deepmind[3]. IBM’s watson, which is used for oncology,
here main task of watson is to suggest patients' medical
health specification and also suggest doctors with
solution option. Apart from these, google'sDeepmind
have achieved positive approach with Alpha Go game
and in near past Deepmind has created Deepmind health
specially for health related terminology.
Generally, bioinformatics domain raw data cannot be
applied to traditional machine-learning techniques [1].
Because traditional techniques requires feature vectors as
a input and particularly for medical imaging analysis have
some issues of coping with variation in imaging
protocols, learning from weak labels, interpretation
and evaluation of results.
Fig. 1: Motivation for deep learning in bioinformatics
For an example to Neural Network (NN) needs a feature
vector as an input. Raw data cannot be used as feature
vector. A domain expert and careful engineering is
needed to process a raw data into a feature vector needed
as input to NN. This preprocessing is expensive and
time consuming task. Deep Neural Network (DNN)
which is one of the DL techniques helps here. DNN
extract features by themselves with multiple hidden
layers. This also called representation learning [2], i.e.,
learning representations of the data that make it easier to
extract useful information when building classifiers or
other predictors. There are mainly three categories of DL
architectures namely Deep Neural Network, Convolution
Neural Network and Recurrent Neural Network. In our
work we have analyzed these DL architectures, problems
needed to be addressed while applying these DL
architectures to bioinformatics problems. Rest of paper is
organized as follows. Deep Neural Network (DNN) has
different architectures depend on which learning
algorithm is used supervised or unsupervised. Based on
this there are mainly three DNN architectures Multilayer
Perceptron (MLP), Deep Belief Network (DBN), Stacked
Auto-Encoder (SAE). As we have discussed earlier
bioinformatics is good domain to explore DL phenomena.
Merriam-Webster has defined Bioinformatics as the
International Journal of Computer Science and Information Security (IJCSIS),
Vol. 15, No. 7, July 2017
76
https://sites.google.com/site/ijcsis/
ISSN 1947-5500
剩余10页未读，继续阅读
资源评论
