Journal of Biomedical Nanotechnology,
Год журнала:
2024,
Номер
20(12), С. 1804 - 1823
Опубликована: Дек. 1, 2024
Predicting
Protein–Protein
Interactions
(PPIs)
is
essential
to
comprehending
biological
functions
and
pivotal
for
drug
discovery
disease
understanding.However,
accurately
predicting
these
interactions
remains
a
difficult
issue
because
of
the
intricate
multifaceted
nature
protein
networks.
Traditional
models
often
fail
fully
capture
relationships
between
proteins
their
interactions,
especially
when
diverse
datasets
are
involved.
To
address
challenges,
novel
approach,
named
Deep
Radial
Graph
Basis
Prism
Refraction
Search
Convolutional
Networks(DRGB-PRSCN)
model,
proposed
PPI
prediction
using
three
distinct
datasets:
Human
PPI,
STRING,
DIP.The
method
employs
Gradient
Domain
Guided
Filtering
effective
data
preprocessing,
ensuring
noise
reduction
while
preserving
features.
Feature
extraction
carried
out
an
Elastic
Decision
Transformer,
which
effectively
captures
key
Networks
(DGCNs)
leveraged
model
complex
dependencies
among
proteins.
The
DRGB-PRSCN
with
its
advanced
architecture,
employed
predict
high
precision.
achieves
performance
evaluation
score
99.9%,
demonstrating
efficacy
in
PPI.
This
approach
outperforms
traditional
methods
by
providing
superior
accuracy
robustness,
making
it
highly
beneficial
network
analysis
discovery.
model’s
primary
benefit
capacity
efficiently
handle
PPIs
exceptional
Biomolecules,
Год журнала:
2025,
Номер
15(1), С. 141 - 141
Опубликована: Янв. 17, 2025
Protein-Protein
Interaction
(PPI)
prediction
plays
a
pivotal
role
in
understanding
cellular
processes
and
uncovering
molecular
mechanisms
underlying
health
disease.
Structure-based
PPI
has
emerged
as
robust
alternative
to
sequence-based
methods,
offering
greater
biological
accuracy
by
integrating
three-dimensional
spatial
biochemical
features.
This
work
summarizes
the
recent
advances
computational
approaches
leveraging
protein
structure
information
for
prediction,
focusing
on
machine
learning
(ML)
deep
(DL)
techniques.
These
methods
not
only
improve
predictive
but
also
provide
insights
into
functional
sites,
such
binding
catalytic
residues.
However,
challenges
limited
high-resolution
structural
data
need
effective
negative
sampling
persist.
Through
integration
of
experimental
tools,
structure-based
paves
way
comprehensive
proteomic
network
analysis,
holding
promise
advancements
drug
discovery,
biomarker
identification,
personalized
medicine.
Future
directions
include
enhancing
scalability
dataset
reliability
expand
these
across
diverse
proteomes.
Signal Transduction and Targeted Therapy,
Год журнала:
2025,
Номер
10(1)
Опубликована: Янв. 6, 2025
Cells
orchestrate
their
processes
through
complex
interactions,
precisely
organizing
biomolecules
in
space
and
time.
Recent
discoveries
have
highlighted
the
crucial
role
of
biomolecular
condensates-membrane-less
assemblies
formed
condensation
proteins,
nucleic
acids,
other
molecules-in
driving
efficient
dynamic
cellular
processes.
These
condensates
are
integral
to
various
physiological
functions,
such
as
gene
expression
intracellular
signal
transduction,
enabling
rapid
finely
tuned
responses.
Their
ability
regulate
signaling
pathways
is
particularly
significant,
it
requires
a
careful
balance
between
flexibility
precision.
Disruption
this
can
lead
pathological
conditions,
including
neurodegenerative
diseases,
cancer,
viral
infections.
Consequently,
emerged
promising
therapeutic
targets,
with
potential
offer
novel
approaches
disease
treatment.
In
review,
we
present
recent
insights
into
regulatory
mechanisms
by
which
influence
pathways,
roles
health
disease,
strategies
for
modulating
condensate
dynamics
approach.
Understanding
these
emerging
principles
may
provide
valuable
directions
developing
effective
treatments
targeting
aberrant
behavior
diseases.
Applied Sciences,
Год журнала:
2025,
Номер
15(6), С. 3283 - 3283
Опубликована: Март 17, 2025
The
prediction
of
protein–protein
interactions
is
a
key
task
in
proteomics.
Since
protein
sequences
are
easily
available
and
understandable,
they
have
become
the
primary
data
source
for
predicting
interactions.
With
development
natural
language
processing
technology,
models
research
hotspot
recent
years,
also
been
developed
accordingly.
Compared
with
single-encoding
methods,
such
as
Word2Vec
one-hot,
specifically
designed
proteins
expected
to
extract
more
comprehensive
information
from
sequences,
thereby
enhancing
performance
interaction
methods.
Inspired
by
model
ProteinBERT,
this
study
LPBERT
deep
learning
framework,
which
novel
end-to-end
architecture.
LPBERT,
based
on
combines
Convolutional
Neural
Networks,
Transformer
encoders,
Bidirectional
Long
Short-Term
Memory
networks
achieve
efficient
prediction.
Upon
evaluation
using
BioGRID
H.
sapiens
S.
cerevisiae
datasets,
outperformed
other
comparison
where
it
achieved
accuracies
98.93%
97.94%,
respectively.
Moreover,
demonstrated
good
performances
multiple
datasets.
These
experimental
results
indicate
that
performed
excellently
tasks,
substantiating
effectiveness
introducing
field.
Computational and Structural Biotechnology Journal,
Год журнала:
2025,
Номер
unknown
Опубликована: Апрель 1, 2025
Assessing
the
potential
toxicity
of
proteins
is
crucial
for
both
therapeutic
and
agricultural
applications.
Traditional
experimental
methods
protein
evaluation
are
time-consuming,
expensive,
labor-intensive,
highlighting
requirement
efficient
computational
approaches.
Recent
advancements
in
language
models
deep
learning
have
significantly
improved
prediction,
yet
current
often
lack
ability
to
integrate
evolutionary
structural
information,
which
accurate
assessment
proteins.
In
this
study,
we
present
ToxDL
2.0,
a
novel
multimodal
model
prediction
that
integrates
information
derived
from
pretrained
AlphaFold2.
2.0
consists
three
key
modules:
(1)
Graph
Convolutional
Network
(GCN)
module
generating
graph
embeddings
based
on
AlphaFold2-predicted
structures,
(2)
domain
embedding
capturing
representations,
(3)
dense
combines
these
predict
toxicity.
After
constructing
comprehensive
benchmark
dataset,
obtained
results
an
original
non-redundant
test
set
(comprising
pre-2022
sequences)
independent
(a
holdout
post-2022
sequences),
demonstrating
outperforms
existing
state-of-the-art
methods.
Additionally,
utilized
Integrated
Gradients
discover
known
toxic
motifs
associated
with
A
web
server
publicly
available
at
www.csbio.sjtu.edu.cn/bioinf/ToxDL2/.
Pharmaceuticals,
Год журнала:
2025,
Номер
18(6), С. 788 - 788
Опубликована: Май 25, 2025
Artificial
intelligence
(AI)
is
a
subfield
of
computer
science
focused
on
developing
systems
that
can
execute
tasks
traditionally
associated
with
human
intelligence.
AI
work
through
algorithms
based
rules
or
instructions
enable
the
machine
to
make
decisions.
With
advancement
science,
more
sophisticated
techniques,
such
as
learning
and
deep
learning,
have
been
developed,
allowing
machines
learn
from
large
amounts
data
improve
their
performance
over
time.
The
pharmaceutical
industry
has
greatly
benefited
development
this
technology.
revolutionized
drug
discovery
by
enabling
rapid
effective
analysis
vast
volumes
biological
chemical
during
identification
new
therapeutic
compounds.
developed
predict
efficacy,
toxicity,
possible
adverse
effects
drugs,
optimize
steps
involved
in
clinical
trials,
reduce
time
costs,
facilitate
implementation
innovative
drugs
market,
making
it
easier
develop
precise
therapies
tailored
individual
genetic
profile
patients.
Despite
significant
advancements,
there
are
still
gaps
application
AI,
particularly
due
lack
comprehensive
regulation.
constant
evolution
technology
requires
ongoing
in-depth
legislative
oversight
ensure
its
use
remains
safe,
ethical,
free
bias.
This
review
explores
role
development,
assessing
potential
enhance
formulation,
accelerate
discovery,
repurpose
existing
medications.
It
highlights
AI’s
impact
across
all
stages,
initial
research
emphasizing
ability
processes,
drive
innovation,
outcomes.
bioRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Май 20, 2024
Drug
discovery
starts
with
known
function,
either
of
a
compound
or
protein,
in-turn
prompting
investigations
to
probe
3D
structure
the
compound-protein
interface.
As
protein
determines
we
hypothesized
that
unique
structural
motifs
represent
primary
information
denoting
function
can
drive
novel
agents.
Using
physics-based
analysis
platform
developed
by
us,
designed
conduct
computationally
intensive
at
supercomputing
speeds,
probed
high-resolution
x-ray
crystallographic
library
us.
We
selected
whose
was
not
otherwise
established,
offered
environments
supporting
binding
drug-like
chemicals
and
were
present
on
proteins
established
therapeutic
targets.
For
each
eight
potential
pockets
six
different
accessed
60
million
used
our
evaluate
binding.
eight-day
colony
formation
assays
acquired
compounds
screened
for
efficacy
against
human
breast,
prostate,
colon
lung
cancer
cells
toxicity
bone
marrow
stem
cells.
Compounds
selectively
inhibiting
growth
segregated
two
separate
proteins.
The
compound,
Dxr2-017,
exhibited
selective
activity
melanoma
in
NCI-60
cell
line
screen,
had
an
IC50
19
nM
M14
assay,
while
over
2100-fold
higher
concentrations
inhibited
less
than
30%.
show
Dxr2-017
induces
anoikis,
form
programmed
death
need
targeted
therapeutics.
predicted
target
is
expressed
bacteria,
humans.
This
supports
strategy
focusing
motifs.
It
functionally
important
structures
are
evolutionarily
conserved.
Here
demonstrate
proof-of-concept
represents
high
value
data
support
approach
widely
applicable.