Research Statement

wickedshortpumpΒιοτεχνολογία

1 Οκτ 2013 (πριν από 3 χρόνια και 10 μήνες)

124 εμφανίσεις

Research

Contributions


Prof. Sanghamitra Bandyopadhyay has
contributed

in the
forefront
areas of
Evolutionary Computation,

P
attern
R
ecognition

and
Bioinformatics
.

She has developed genetic algorithm based classifiers by appropriately modeling
the class bo
undaries using a number of hyperplanes. Theoretical analysis of the genetic classifier establishes that
its performances approaches that of the Bayes classifier under limiting conditions.
Evolutionary clustering is an
important area of research that focuse
s on the interplay of computational intelligence and pattern recognition.
T
he existing approaches
in this domain

used
data point
based encoding of the solutions, and hence suffered from
scalability issues. Sanghamitra’s pioneering proposal of
cluster cente
r

based encoding
led to faster convergence

of
the techniques
, and enabled their
application to large data sets.
Through

in
-
depth study of evolutionary clustering,
she

developed new, theoretically sound, measures of cluster validity and stability
.
Her contr
ibutions have made
evolutionary clustering applicable

to a

broad spectrum

of data

with widely varying characteristics
.
The applied
aspect of the
her

work is demonstrated by the successful applications of her techniques
to
many real
-
life problems
like
analo
g modulation classification, lip segmentation, biomedical and satellite image segmentation, in the
CANARY software for
water
quality event detection, modeling adiabatic temperature rise during concrete
hydration, biometric analysis, reconstruction of gene
regulatory network, and bioinformatics.


Prof. Bandyopadhyay is one of
the first few researchers

to pose the clustering problem in a multiobjective
framework
so as to identify
multiple but relevant partitionings of the data.

She proposed an innovative str
ategy of
integrating multiobjective clustering with supervised learning for combining the multiple
Pareto
-
optimal
solutions
resulting from the former in order to evolve a single superior solution.
She has
co
-
authored
a

book
title
d

Multiobjective Genetic Al
gorithms for Clustering: Applications in Data Mining and Bioinformatics
in 2011,
the
first

such volume

containing an in
-
depth discussion on different aspects of multiobjective clustering.
She has also
proposed a new multiobjective simulated annealing metho
d, AMOSA, which is found to outperform several
existing widely used multiobjective metaheuristics especially for a large number of objectives. The main novelty in
AMOSA is the following
-

in contrast to existing techniques which always accept a better solu
tion when compared
to an inferior one, in AMOSA the inferior solution has a non
-
zero chance of survival. This makes AMOSA less
greedy in nature, thereby providing improved performance in the long run.


MicroRNAs are small non
-
coding RNA molecules
that
reg
ulate other genes post
-
transcriptional
ly
, inhibiting the
production of
the
proteins.
Abnormal microRNA levels
are

observed in many different diseases including almost all
types of

cancer, indicating that these molecules play critical roles in disease progr
ession
.
Prof. Bandyopadhyay has,
for the first time, developed a cancer
-
microRNA network whose analysis revealed useful insights and raised
interesting biological questions.
It is used to identify hub miRNAs and miRNAs with strong onco/tumor suppressor
cha
racteristics.

The paper appeared as a featured article in BiomedCentral website, life
-
sciences newsletter
eBioNews (
http://www.ebionews.com/news
-
center/research
-
frontiers/rnai
-
a
-
microrna/14853
-
development
-
of
-
the
-
human
-
cancer
-
microrna
-
network.html
),

and in the annual cancer issue of Genome Technology, April 2010.
She
led the development of a comprehensive miRNA
-
T
ranscription Factor

(TF)
-
gene regulatory network to infer crucial
secondary miRNA regulations, inc
luding miRNA
-
miRNA regulations
.


Development of TargetMiner, machine learning based approach for predicting microRNA targets, is another
significant contribu
tion of Sanghamitra. A major impediment here was the lack of negative
data

for training. Earlier
prediction methods used randomly selected/artificially generated negative samples, and suffered from high false
positive rates. In contrast,
in
TargetMiner
bio
logically relevant, tissue
-
specific negative samples were
systematically identified. Consequently, it provided superior specificity
-
sensitivity trade
-
off as compared to the
existing methods.
The work featured in BioTech Wee
k, Nov. 4,
2009, and the
web base
d tool

TargetMiner
(
www.isical.ac.in/~bioinfo_miu
), generated significant interest in the research community with the
site

getting
more than 1000 hits within the first four months of its launch.
It is
al
so
included in Canada’s Michael Smith
Genome Sciences Center


(
http://www.bcgsc.ca/downloads/genereg/remcbigdata/miR/TargetMiner/
).
She has
also developed a series of rank
aggregation methods, one of which has been selected for filing of patent by
Intellectual Ventures. She has applied these techniques for aggregating the microRNA target rankings provided by
multiple target prediction methods. In fact the developed technique
s can also be applied in the domain of
webmining. Besides these, Prof. Bandyopdhyay has contributed extensively in the domains of microarray data
clustering, yeast gene function prediction,
finding strongly associated modules in coexpression graphs, etc.