2009 DrosophilaGeneExpressionPatternAnnotation Abstract

See: Abstract Analysis Report, (Ji et al., 2009) ⇒ Shuiwang Ji, Lei Yuan, Ying-Xin Li, Zhi-Hua Zhou, Sudhir Kumar, and Jieping Ye. (2009). “Drosophila Gene Expression Pattern Annotation Using Sparse Features and Term-term Interactions.” In: Proceedings of ACM SIGKDD Conference (KDD-2009).

---

The	The
Drosophila	Drosophila
gene	gene expression pattern	Gene Expression Pattern
expression
pattern
images	images
document	document
the	the
spatial	spatial
and	and
temporal	temporal
dynamics	dynamics
of	of
gene	gene expression	Gene Expression
expression
and	and
they	they
are	are
valuable	valuable
tools	tools
for	for
explicating	explicating
the	the
gene	gene
functions	functions
,	,
interaction	interaction
,	,
and	and
networks	networks
during	during
Drosophila	Drosophila
embryogenesis	embryogenesis
.	.
To	To
provide	provide
text-based	text-based
pattern	pattern
searching	searching
,	,
the	the
images	images
in	in
the	the
Berkeley	Berkeley Drosophila Genome Project	Berkeley Drosophila Genome Project
Drosophila
Genome
Project
(	(
BDGP	BDGP
)	)
study	study
are	are
annotated	annotated
with	with
ontology	ontology
terms	terms
manually	manually
by	by
human	human curator
curators
.	.
We	We
present	present
a	a
systematic	systematic
approach	approach
for	for
automating	automating
this	this
task	task
,	,
because	because
the	the
number	number
of	of
images	images
needing	needing
text	text
descriptions	descriptions
is	is
now	now
rapidly	rapidly
increasing	increasing
.	.
We	We
consider	consider
both	both
improved	improved
feature	feature
representation	representation
and	and
novel	novel
learning	learning	?
formulation	formulation
to	to
boost	boost
the	the
annotation	annotation	Annotation Task
performance	performance	Performance
.	.
For	For
feature	feature
representation	representation
,	,
we	we
adapt	adapt
the	the
bag-of-words	bag-of-words
scheme	scheme
commonly	commonly
used	used
in	in
visual	visual recognition problem	Image Recognition Task
recognition
problems
so	so
that	that
the	the
image	image
group	group
information	information
in	in
the	the
BDGP	BDGP	BDGP
study	study
is	is
retained	retained
.	.
Moreover	Moreover
,	,
images	images
from	from
multiple	multiple
views	views
can	can
be	be
integrated	integrated
naturally	naturally
in	in
this	this
representation	representation
.	.
To	To
reduce	reduce
the	the
quantization	quantization
error	error
caused	caused
by	by
the	the
bag-of-words	bag-of-words representation	Bag-of-Words Representation
representation,
,	,
we	we
propose	propose
an	an
improved	improved
feature	feature representation	Feature Representation
representation
scheme	scheme
based	based
on	on
the	the
sparse	sparse learning technique	Sparse Learning Technique
learning
technique.
.	.
In	In
the	the
design	design
of	of
learning	learning
formulation	formulation
,	,
we	we
propose	propose
a	a
local	local
regularization	regularization
framework	framework
that	that
can	can
incorporate	incorporate
the	the
correlations	correlations
among	among
terms	terms
explicitly	explicitly
.	.
We	We
further	further
show	show
that	that
the	the
resulting	resulting
optimization	optimization problem	Optimization Task
problem
admits	admits
an	an
analytical	analytical
solution	solution
.	.
Experimental	Experimental results	[[Experimental Result]]
results
show	show
that	that
the	the
representation	representation
based	based
on	on
sparse	sparse learning	Sparse Learning
learning
outperforms	outperforms
the	the
bag-of-words	bag-of-words	Bag-of-Words Representation
representation	representation
significantly	significantly
.	.
Results	Results	Experimental Result
also	also
show	show
that	that
incorporation	incorporation
of	of
the	the
term-term	term-term	Term-Term Relation
correlations	correlations	Correlation
improves	improves
the	the
annotation	annotation	Annotation Task
performance	performance	Performance
consistently	consistently
.	.

,