bioRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2025,
Volume and Issue:
unknown
Published: April 19, 2025
A
bstract
Predicting
synergistic
cancer
drug
combinations
through
computational
methods
offers
a
scalable
approach
to
creating
therapies
that
are
more
effective
and
less
toxic.
However,
most
algorithms
focus
solely
on
synergy
without
considering
toxicity
when
selecting
optimal
combinations.
In
the
absence
of
combinatorial
assays,
few
models
use
penalties
balance
high
with
lower
toxicity.
these
have
not
been
explicitly
validated
against
known
drug-drug
interactions.
this
study,
we
examine
whether
scores
metrics
correlate
adverse
While
some
show
trends
levels,
our
results
reveal
significant
limitations
in
using
them
as
penalties.
These
findings
highlight
challenges
incorporating
into
prediction
frameworks
suggest
advancing
field
requires
comprehensive
combination
data.
Briefings in Bioinformatics,
Journal Year:
2023,
Volume and Issue:
25(1)
Published: Nov. 22, 2023
Abstract
Recently,
attention
mechanism
and
derived
models
have
gained
significant
traction
in
drug
development
due
to
their
outstanding
performance
interpretability
handling
complex
data
structures.
This
review
offers
an
in-depth
exploration
of
the
principles
underlying
attention-based
advantages
discovery.
We
further
elaborate
on
applications
various
aspects
development,
from
molecular
screening
target
binding
property
prediction
molecule
generation.
Finally,
we
discuss
current
challenges
faced
application
mechanisms
Artificial
Intelligence
technologies,
including
quality,
model
computational
resource
constraints,
along
with
future
directions
for
research.
Given
accelerating
pace
technological
advancement,
believe
that
will
increasingly
prominent
role
anticipate
these
usher
revolutionary
breakthroughs
pharmaceutical
domain,
significantly
development.
Frontiers in Pharmacology,
Journal Year:
2024,
Volume and Issue:
15
Published: Feb. 7, 2024
There
are
two
main
ways
to
discover
or
design
small
drug
molecules.
The
first
involves
fine-tuning
existing
molecules
commercially
successful
drugs
through
quantitative
structure-activity
relationships
and
virtual
screening.
second
approach
generating
new
de
novo
inverse
relationship.
Both
methods
aim
get
a
molecule
with
the
best
pharmacokinetic
pharmacodynamic
profiles.
However,
bringing
market
is
an
expensive
time-consuming
endeavor,
average
cost
being
estimated
at
around
$2.5
billion.
One
of
biggest
challenges
screening
vast
number
potential
candidates
find
one
that
both
safe
effective.
development
artificial
intelligence
in
recent
years
has
been
phenomenal,
ushering
revolution
many
fields.
field
pharmaceutical
sciences
also
significantly
benefited
from
multiple
applications
intelligence,
especially
discovery
projects.
Artificial
models
finding
use
molecular
property
prediction,
generation,
screening,
synthesis
planning,
repurposing,
among
others.
Lately,
generative
gained
popularity
across
domains
for
its
ability
generate
entirely
data,
such
as
images,
sentences,
audios,
videos,
novel
chemical
molecules,
etc.
Generative
delivered
promising
results
development.
This
review
article
delves
into
fundamentals
framework
various
context
via
approach.
Various
basic
advanced
have
discussed,
along
their
applications.
explores
examples
advances
approach,
well
ongoing
efforts
fully
harness
faster
more
affordable
manner.
Some
clinical-level
assets
generated
form
discussed
this
show
ever-increasing
application
commercial
partnerships.
Journal of Chemical Information and Modeling,
Journal Year:
2024,
Volume and Issue:
64(7), P. 2577 - 2585
Published: March 22, 2024
Drug
synergy
prediction
plays
a
vital
role
in
cancer
treatment.
Because
experimental
approaches
are
labor-intensive
and
expensive,
computational-based
get
more
attention.
There
two
types
of
computational
methods
for
drug
prediction:
feature-based
similarity-based.
In
methods,
the
main
focus
is
to
extract
discriminative
features
from
pairs
cell
lines
pass
task
predictor.
similarity-based
similarities
among
all
drugs
utilized
as
fed
into
this
work,
novel
approach,
called
CFSSynergy,
that
combines
these
viewpoints
proposed.
First,
representation
extracted
paired
input.
We
have
transformer-based
architecture
drugs.
For
lines,
we
created
similarity
matrix
between
proteins
using
Node2Vec
algorithm.
Then,
new
line
computed
by
multiplying
protein–protein
initial
representation.
Next,
compute
unique
cells
learned
lines.
based
on
features.
Finally,
XGBoost
Two
well-known
data
sets
were
used
evaluate
performance
our
proposed
method:
DrugCombDB
OncologyScreen.
The
CFSSynergy
approach
consistently
outperformed
existing
comparative
evaluations.
This
substantiates
efficacy
capturing
complex
synergistic
interactions
setting
it
apart
conventional
or
methods.
Briefings in Bioinformatics,
Journal Year:
2024,
Volume and Issue:
25(2)
Published: Jan. 22, 2024
Abstract
Spatial
transcriptomics
technologies
have
shed
light
on
the
complexities
of
tissue
structures
by
accurately
mapping
spatial
microenvironments.
Nonetheless,
a
myriad
methods,
especially
those
utilized
in
platforms
like
Visium,
often
relinquish
details
owing
to
intrinsic
resolution
limitations.
In
response,
we
introduce
TransformerST,
an
innovative,
unsupervised
model
anchored
Transformer
architecture,
which
operates
independently
references,
thereby
ensuring
cost-efficiency
circumventing
need
for
single-cell
RNA
sequencing.
TransformerST
not
only
elevates
Visium
data
from
multicellular
level
granularity
but
also
showcases
adaptability
across
diverse
platforms.
By
employing
vision
transformer-based
encoder,
it
discerns
latent
image-gene
expression
co-representations
and
is
further
enhanced
correlations,
derived
adaptive
graph
module.
The
sophisticated
cross-scale
network,
super-resolution,
significantly
boosts
model’s
accuracy,
unveiling
complex
structure–functional
relationships
within
histology
images.
Empirical
evaluations
validate
its
adeptness
revealing
subtleties
at
scale.
Crucially,
adeptly
navigates
through
co-representation,
maximizing
synergistic
utility
gene
images,
emerging
as
pioneering
tool
transcriptomics.
It
enhances
introduces
novel
approach
that
optimally
utilizes
images
alongside
expression,
providing
refined
lens
investigating
Journal of Biomolecular Structure and Dynamics,
Journal Year:
2023,
Volume and Issue:
unknown, P. 1 - 10
Published: Dec. 12, 2023
Virtual
screening
has
emerged
as
a
valuable
computational
tool
for
predicting
compound-protein
interactions,
offering
cost-effective
and
rapid
approach
to
identifying
potential
candidate
drug
molecules.
Current
machine
learning-based
methods
rely
on
molecular
structures
their
relationship
in
the
network.
The
former
utilizes
information
such
amino
acid
sequences
chemical
structures,
while
latter
leverages
interaction
network
data,
protein-protein
drug-disease
protein-disease
interactions.
However,
there
been
limited
exploration
of
integrating
with
networks.
This
study
presents
DeepCompoundNet,
deep
model
that
integrates
protein
features,
properties,
diverse
data
predict
chemical-protein
DeepCompoundNet
outperforms
state-of-the-art
prediction,
demonstrated
through
performance
evaluations.
Our
findings
highlight
complementary
nature
multiple
extending
beyond
sequence
homology
structure
similarity.
Moreover,
our
model's
analysis
confirms
gets
higher
interactions
between
proteins
chemicals
not
observed
training
samples.Communicated
by
Ramaswamy
H.
Sarma.
IEEE Journal of Biomedical and Health Informatics,
Journal Year:
2024,
Volume and Issue:
28(4), P. 2445 - 2454
Published: Jan. 4, 2024
Origins
of
replication
sites
(ORIs)
are
crucial
genomic
regions
where
DNA
initiation
takes
place,
playing
pivotal
roles
in
fundamental
biological
processes
like
cell
division,
gene
expression
regulation,
and
integrity.
Accurate
identification
ORIs
is
essential
for
comprehending
replication,
expression,
mutation-related
diseases.
However,
experimental
approaches
ORI
often
expensive
time-consuming,
leading
to
the
growing
popularity
computational
methods.
In
this
study,
we
present
PLANNER
(DeeP
LeArNiNg
prEdictor
ORI),
a
novel
approach
species-specific
cell-specific
prediction
eukaryotic
ORIs.
uses
multi-scale
k-tuple
sequences
as
input
employs
DNABERT
pre-training
model
with
transfer
learning
ensemble
strategies
train
accurate
predictive
models.
Extensive
empirical
test
results
demonstrate
that
achieved
superior
performance
compared
state-of-the-art
approaches,
including
iOri-Euk,
Stack-ORI,
ORI-Deep,
within
specific
types
across
different
types.
Furthermore,
by
incorporating
an
interpretable
analysis
mechanism,
provide
insights
into
learned
patterns,
facilitating
mapping
from
discovering
important
sequential
determinants
comprehensively
analysing
their
functions.
Briefings in Bioinformatics,
Journal Year:
2024,
Volume and Issue:
25(2)
Published: Jan. 22, 2024
Abstract
Accurate
cell
type
annotation
in
single-cell
RNA-sequencing
data
is
essential
for
advancing
biological
and
medical
research,
particularly
understanding
disease
progression
tumor
microenvironments.
However,
existing
methods
are
constrained
by
single
feature
extraction
approaches,
lack
of
adaptability
to
immune
types
with
similar
molecular
profiles
but
distinct
functions
a
failure
account
the
impact
label
noise
on
model
accuracy,
all
which
compromise
precision
annotation.
To
address
these
challenges,
we
developed
supervised
approach
called
scMMT.
We
proposed
novel
technique
uncover
more
valuable
information.
Additionally,
constructed
multi-task
learning
framework
based
GradNorm
method
enhance
recognition
challenging
cells
reduce
facilitating
mutual
reinforcement
between
protein
prediction
tasks.
Furthermore,
introduced
logarithmic
weighting
smoothing
mechanisms
ability
rare
prevent
overconfidence.
Through
comprehensive
evaluations
multiple
public
datasets,
scMMT
has
demonstrated
state-of-the-art
performance
various
aspects
including
annotation,
identification,
dropout
resistance,
expression
low-dimensional
embedding
representation.