bbc 2015

BeNeLux Bioinformatics Conference – Antwerp, December 7-8 2015
Abstract ID: P
Poster
10th Benelux Bioinformatics Conference bbc 2015
P59. MAPREDUCE APPROACHES FOR CONTACT MAP PREDICTION:
AN EXTREMELY IMBALANCED BIG DATA PROBLEM
Isaac Triguero 1,2* , Sara del Río 3 , Victoria López 3 , Jaume Bacardit 4 , José M. Benítez 3 & Francisco Herrera 3 .
VIB Inflammation Research Center 1 ; Department of Respiratory Medicine, Ghent University 2 ; Department of Computer
Science and Artificial Intelligence 3 ; School of Computing Science, Newcastle University 4 .
* Isaac.Triguero@irc.vib-Ugent.be
The application of data mining and machine learning techniques to biological and biomedicine data continues to be an
ubiquitous research theme in current bioinformatics. The rapid advances in biotechnology are allowing us to obtain and
store large quantities of data about cells, proteins, genes, etc, that should be processed. Moreover, in many of these
problems such as contact map prediction, it is difficult to collect representative positive examples. Learning under these
circumstances, known as imbalance big data classification, may not be straightforward for most of the standard machine
learning methods. In this work we describe the methodology that won the ECBDL'14 big data competition, which was
concerned with the prediction of contact maps. Our methodology is composed of several MapReduce approaches to deal
with big amounts of data. The results show that this model is very suitable to tackle large-scale bioinformatics
classifications problems.
INTRODUCTION
The prediction of a protein’s contact map is a crucial step
for the prediction of the complete 3D structure of a protein.
This is one of the most challenging bioinformatics tasks
within the field of protein structure prediction because of
the sparseness of the contacts (i.e. few positive examples)
and the great amount of data extracted (i.e. millions of
instances, Gbs of disk space) from a few thousand of
proteins.
This problem refers to an imbalance bioinformatics big
data application, in which traditional machine learning
techniques become non effective and non efficient due to
the big dimension of the problem. However, with use of
the emerging cloud-based technologies, these techniques
can be redesigned to extract valuable knowledge from
such amount of data.
The ECDBL’14 competition (http://cruncher.ncl.ac.uk/
bdcomp/) brought up a data set that modeled the contact
map prediction problem as a classification task.
Concretely, the training data set considered was formed by
32 million instances, 631 attributes, 2 classes, 98% of
negative examples and it occupies about 56GB of disk
space.
In this work we describe the methodology with which we
have participated, under the name 'Efdamis', ranking as the
winner algorithm (Triguero et al, 2015).
METHODS
In the proposed methodology, we focused on the
MapReduce (Dean et al, 2008) paradigm in order to
manage this voluminous data set. We extended the
applicability of some pre-processing and classification
models to deal with large-scale problems. This is
composed of four main parts:
An oversampling approach: The goal is to balance the
highly skewed class distribution of the problem by
replicating randomly the instances of the minority
class (del Rio et al, 2014).
An evolutionary feature weighting method: Due the
relative high number of features of the given problem
we developed a feature selection scheme for largescale
problems that improves the classification
performance by detecting the most significant features
(Triguero et al, 2012).
Building a learning model: As classifier, we focused
on a scalable RandomForest algorithm.
Testing the model: Even the test data can be
considered big data (2.9 millions of instances), so that,
the testing phase was also deployed within a parallel
approach.
RESULTS & DISCUSSION
Table 1 presents the final results of the top 5 participants
in terms of True Positive Rate (TPR) and True Negative
Rate (TNR). In this particular problem, the necessity of
balancing the TPR and TNR ratios emerged as a difficult
challenge for most of the participants of the competition.
In this sense, the use of scalable preprocessing techniques
played in important role to improve the results of the
RandomForest classifier. First, the designed oversampling
approach allowed us to prevent RandomForest to be
biased to the negative class. Second, our feature weighting
approach provided us the possibility of reducing the
dimensionality of the problem by selecting the most
relevant features. Thus, it resulted in a better performance
as well as a notable reduction of the time requirements.
Team TPR TNR TPR * TNR
Efdamis 0.73043 0.73018 0.53335
ICOS 0.70321 0.73016 0.51345
UNSW 0.69916 0.72763 0.50873
HyperEns 0.64003 0.76338 0.48858
PUC-Rio_ICA 0.65709 0.71460 0.46956
TABLE 1: Comparison with the top 5 of the competition.
REFERENCES
Dean J., Ghemawat S., Mapreduce: simplified data processing on large
clusters, Commun. ACM 51 (1), 107–113 (2008).
del Río S., et al., On the use of MapReduce for imbalanced big data using
random forest, Inf. Sci. 285 (2014) 112–137.
Triguero I. et al., Integrating a differential evolution feature weighting
scheme into prototype generation, Neurocomputing 97 (2012) 332–
343.
103