Abstract
Motivation
Exploring
potential
associations
between
diseases
can
help
in
understanding
pathological
mechanisms
of
and
facilitating
the
discovery
candidate
biomarkers
drug
targets,
thereby
promoting
disease
diagnosis
treatment.
Some
computational
methods
have
been
proposed
for
measuring
similarity.
However,
these
describe
without
considering
their
latent
multi-molecule
regulation
valuable
supervision
signal,
resulting
limited
biological
interpretability
efficiency
to
capture
association
patterns.
Results
In
this
study,
we
propose
a
new
method
named
DiSMVC.
Different
from
existing
predictors,
DiSMVC
designs
supervised
graph
collaborative
framework
measure
Multiple
bio-entity
related
genes
miRNAs
are
integrated
via
cross-view
contrastive
learning
extract
informative
representation,
then
pattern
joint
is
implemented
compute
similarity
by
incorporating
phenotype-annotated
associations.
The
experimental
results
show
that
draw
discriminative
characteristics
pairs,
outperform
other
state-of-the-art
methods.
As
result,
promising
predicting
with
molecular
interpretability.
Availability
implementation
Datasets
source
codes
available
at
https://github.com/Biohang/DiSMVC.
Journal of Chemical Information and Modeling,
Год журнала:
2025,
Номер
unknown
Опубликована: Янв. 19, 2025
Recent
studies
have
highlighted
the
significant
role
of
circular
RNAs
(circRNAs)
in
various
diseases.
Accurately
predicting
circRNA–disease
associations
is
crucial
for
understanding
their
biological
functions
and
disease
mechanisms.
This
work
introduces
MNDCDA
method,
designed
to
address
challenges
posed
by
limited
number
known
high
cost
experiments.
integrates
multiple
data
sources
with
neighborhood-aware
embedding
models
deep
feature
projection
networks
predict
potential
pathways
linking
circRNAs
Initially,
comprehensive
biometric
are
used
construct
four
similarity
networks,
forming
a
diverse
interaction
framework.
Next,
model
captures
structural
information
about
diseases,
while
learn
high-order
interactions
nonlinear
connections.
Finally,
bilinear
decoder
identifies
novel
between
The
achieved
an
AUC
0.9070
on
constructed
benchmark
dataset.
In
case
studies,
25
out
30
predicted
pairs
were
validated
through
wet
lab
experiments
published
literature.
These
extensive
experimental
results
demonstrate
that
robust
computational
tool
associations,
providing
valuable
insights
helping
reduce
research
costs.
Briefings in Bioinformatics,
Год журнала:
2024,
Номер
25(3)
Опубликована: Март 27, 2024
Abstract
Identifying
disease-associated
microRNAs
(miRNAs)
could
help
understand
the
deep
mechanism
of
diseases,
which
promotes
development
new
medicine.
Recently,
network-based
approaches
have
been
widely
proposed
for
inferring
potential
associations
between
miRNAs
and
diseases.
However,
these
ignore
importance
different
relations
in
meta-paths
when
learning
embeddings
Besides,
they
pay
little
attention
to
screening
out
reliable
negative
samples
is
crucial
improving
prediction
accuracy.
In
this
study,
we
propose
a
novel
approach
named
MGCNSS
with
multi-layer
graph
convolution
high-quality
sample
selection
strategy.
Specifically,
first
constructs
comprehensive
heterogeneous
network
by
integrating
miRNA
disease
similarity
networks
coupled
their
known
association
relationships.
Then,
employ
automatically
capture
meta-path
lengths
learn
discriminative
representations
After
that,
establishes
highly
set
from
unlabeled
distance-based
Finally,
train
under
an
unsupervised
manner
predict
The
experimental
results
fully
demonstrate
that
outperforms
all
baseline
methods
on
both
balanced
imbalanced
datasets.
More
importantly,
conduct
case
studies
colon
neoplasms
esophageal
neoplasms,
further
confirming
ability
detect
candidate
miRNAs.
source
code
publicly
available
GitHub
https://github.com/15136943622/MGCNSS/tree/master
Journal of Chemical Information and Modeling,
Год журнала:
2025,
Номер
unknown
Опубликована: Фев. 19, 2025
Cancer
affects
millions
globally,
and
as
research
advances,
our
understanding
treatment
of
cancer
evolve.
Compared
to
conventional
treatments
with
significant
side
effects,
anticancer
peptides
(ACPs)
have
gained
considerable
attention.
Validating
ACPs
through
wet-lab
experiments
is
time-consuming
costly.
However,
numerous
artificial
intelligence
methods
are
now
used
for
ACP
identification
classification.
These
typically
apply
a
uniform
strategy
all
feature
types,
overlooking
the
potential
benefits
more
specialized
processing
different
types.
In
this
paper,
we
propose
framework
based
on
multichannel
discriminative
processing,
where
neural
networks
applied
process
various
optimizing
their
respective
vectors.
Additionally,
leverage
Large
Pretrained
Protein
Language
Models
capture
deeper
sequence
features,
further
enhancing
model's
performance.
Contributions:
To
better
validate
overall
performance
generalization
ability
model,
compared
it
state-of-the-art
models
using
four
data
sets
(AntiCp2Main,
AntiCp2
Alternate,
ACP740,
cACP-DeepGram).
The
results
show
improvements
across
most
metrics.
proposed
assists
researchers
in
distinguishing
identifying
validates
need
distinct
PLoS Computational Biology,
Год журнала:
2024,
Номер
20(8), С. e1012399 - e1012399
Опубликована: Авг. 22, 2024
Circular
RNAs
(circRNAs)
play
vital
roles
in
transcription
and
translation.
Identification
of
circRNA-RBP
(RNA-binding
protein)
interaction
sites
has
become
a
fundamental
step
molecular
cell
biology.
Deep
learning
(DL)-based
methods
have
been
proposed
to
predict
achieved
impressive
identification
performance.
However,
those
cannot
effectively
capture
long-distance
dependencies,
utilize
the
information
multiple
features.
To
overcome
limitations,
we
propose
DL-based
model
iCRBP-LKHA
using
deep
hybrid
networks
for
identifying
sites.
adopts
five
encoding
schemes.
Meanwhile,
neural
network
architecture,
which
consists
large
kernel
convolutional
(LKCNN),
block
attention
module
with
one-dimensional
convolution
(CBAM-1D)
bidirectional
gating
recurrent
unit
(BiGRU),
can
explore
local
information,
global
context
features
automatically.
verify
effectiveness
iCRBP-LKHA,
compared
its
performance
shallow
algorithms
on
37
circRNAs
datasets
stringent
datasets.
And
state-of-the-art
datasets,
31
linear
The
experimental
results
not
only
show
that
outperforms
other
competing
methods,
but
also
demonstrate
potential
this
RNA-RBP
Briefings in Functional Genomics,
Год журнала:
2025,
Номер
24
Опубликована: Янв. 1, 2025
Abstract
Deep
learning
models
have
made
significant
progress
in
the
biomedical
field,
particularly
prediction
of
drug–drug
interactions
(DDIs).
DDIs
are
pharmacodynamic
reactions
between
two
or
more
drugs
body,
which
may
lead
to
adverse
effects
and
great
significance
for
drug
development
clinical
research.
However,
predicting
DDI
through
traditional
trials
experiments
is
not
only
costly
but
also
time-consuming.
When
utilizing
advanced
Artificial
Intelligence
(AI)
deep
techniques,
both
developers
users
face
multiple
challenges,
including
problem
acquiring
encoding
data,
as
well
difficulty
designing
computational
methods.
In
this
paper,
we
review
a
variety
methods,
similarity-based,
network-based,
integration-based
approaches,
provide
an
up-to-date
easy-to-understand
guide
researchers
different
fields.
Additionally,
in-depth
analysis
widely
used
molecular
representations
systematic
exposition
theoretical
framework
extract
features
from
graph
data.
Drug-disease
association
(DDA)
prediction
aims
to
identify
potential
links
between
drugs
and
diseases,
facilitating
the
discovery
of
new
therapeutic
potentials
reducing
cost
time
associated
with
traditional
drug
development.
However,
existing
DDA
methods
often
overlook
global
relational
information
provided
by
other
biological
entities,
complex
structure
limiting
correlations
disease
embeddings.
In
this
study,
we
propose
HNF-DDA,
a
subgraph
contrastive-driven
transformer-style
heterogeneous
network
embedding
model
for
prediction.
Specifically,
HNF-DDA
adopts
all-pairs
message
passing
strategy
capture
network,
fully
integrating
multi-omics
information.
also
proposes
concept
contrastive
learning
local
drug-disease
subgraphs,
high-order
semantic
nodes.
Experimental
results
on
two
benchmark
datasets
demonstrate
that
outperforms
several
state-of-the-art
methods.
Additionally,
it
shows
superior
performance
across
different
dataset
splitting
schemes,
indicating
HNF-DDA's
capability
generalize
novel
categories.
Case
studies
breast
cancer
prostate
reveal
9
out
top
10
predicted
candidate
8
have
documented
effects.
incorporates
strategies
into
embedding,
enabling
effective
representations
enriched
information,
while
demonstrating
significant
applications
in
repositioning.
Three-dimensional
molecular
generation
is
critical
in
drug
design.
However,
current
methods
often
rely
on
point
clouds
or
oversimplified
interaction
models,
limiting
their
ability
to
accurately
represent
structures.
To
address
these
challenges,
this
paper
proposes
the
multiscale
graph
equivariant
diffusion
model
for
3D
molecule
design
(MD3MD).
MD3MD
partitions
conformations
into
graphs,
assigning
different
weights
capture
atomic
interactions
across
scales.
This
framework
guides
process,
enabling
high-quality
generation.
Experimental
results
demonstrate
that
excels
both
unconditional
and
conditional
tasks,
producing
diverse,
stable,
innovative
molecules
meet
specified
conditions.
Visualization
highlights
MD3MD’s
learn
domain-specific
patterns
generate
distinct
from
existing
datasets
while
maintaining
distributional
consistency.
By
effectively
exploring
chemical
space,
surpasses
previous
generating
chemically
diverse
molecules,
offering
a
notable
advancement
field
of
Abstract
Background
Numerous
studies
have
shown
that
circRNA
can
act
as
a
miRNA
sponge,
competitively
binding
to
miRNAs,
thereby
regulating
gene
expression
and
disease
progression.
Due
the
high
cost
time-consuming
nature
of
traditional
wet
lab
experiments,
analyzing
circRNA-miRNA
associations
is
often
inefficient
labor-intensive.
Although
some
computational
models
been
developed
identify
these
associations,
they
fail
capture
deep
collaborative
features
between
interactions
do
not
guide
training
feature
extraction
networks
based
on
high-order
relationships,
leading
poor
prediction
performance.
Results
To
address
issues,
we
innovatively
propose
novel
graph
collaboration
learning
method
for
interaction,
called
DGCLCMI.
First,
it
uses
word2vec
encode
sequences
into
word
embeddings.
Next,
present
joint
model
combines
an
improved
neural
filtering
with
network
optimization.
Deep
interaction
information
embedded
informative
within
sequence
representations
prediction.
Comprehensive
experiments
three
well-established
datasets
across
seven
metrics
demonstrate
our
algorithm
significantly
outperforms
previous
models,
achieving
average
AUC
0.960.
In
addition,
case
study
reveals
18
out
20
predicted
unknown
CMI
data
points
are
accurate.
Conclusions
The
DGCLCMI
improves
representation
by
capturing
information,
superior
performance
compared
prior
methods.
It
facilitates
discovery
sheds
light
their
roles
in
physiological
processes.