Application of Transformers in Cheminformatics
Journal of Chemical Information and Modeling,
Год журнала:
2024,
Номер
64(11), С. 4392 - 4409
Опубликована: Май 30, 2024
By
accelerating
time-consuming
processes
with
high
efficiency,
computing
has
become
an
essential
part
of
many
modern
chemical
pipelines.
Machine
learning
is
a
class
methods
that
can
discover
patterns
within
data
and
utilize
this
knowledge
for
wide
variety
downstream
tasks,
such
as
property
prediction
or
substance
generation.
The
complex
diverse
space
requires
machine
architectures
great
power.
Recently,
models
based
on
transformer
have
revolutionized
multiple
domains
learning,
including
natural
language
processing
computer
vision.
Naturally,
there
been
ongoing
endeavors
in
adopting
these
techniques
to
the
domain,
resulting
surge
publications
short
period.
diversity
structures,
use
cases,
necessitate
comprehensive
summarization
existing
works.
In
paper,
we
review
recent
innovations
adapting
transformers
solve
problems
chemistry.
Because
complex,
structure
our
discussion
representations.
Specifically,
highlight
strengths
weaknesses
each
representation,
current
progress
architectures,
future
directions.
Язык: Английский
Research on Bitter Peptides in the Field of Bioinformatics: A Comprehensive Review
International Journal of Molecular Sciences,
Год журнала:
2024,
Номер
25(18), С. 9844 - 9844
Опубликована: Сен. 12, 2024
Bitter
peptides
are
small
molecular
produced
by
the
hydrolysis
of
proteins
under
acidic,
alkaline,
or
enzymatic
conditions.
These
can
enhance
food
flavor
and
offer
various
health
benefits,
with
attributes
such
as
antihypertensive,
antidiabetic,
antioxidant,
antibacterial,
immune-regulating
properties.
They
show
significant
potential
in
development
functional
foods
prevention
treatment
diseases.
This
review
introduces
diverse
sources
bitter
discusses
mechanisms
bitterness
generation
their
physiological
functions
taste
system.
Additionally,
it
emphasizes
application
bioinformatics
peptide
research,
including
establishment
improvement
databases,
use
quantitative
structure–activity
relationship
(QSAR)
models
to
predict
thresholds,
latest
advancements
classification
prediction
built
using
machine
learning
deep
algorithms
for
identification.
Future
research
directions
include
enhancing
diversifying
models,
applying
generative
advance
towards
deepening
discovering
more
practical
applications.
Язык: Английский
Computational methods for modeling protein–protein interactions in the AI era: Current status and future directions
Drug Discovery Today,
Год журнала:
2025,
Номер
unknown, С. 104382 - 104382
Опубликована: Май 1, 2025
Язык: Английский
Predicting mutation-disease associations through protein interactions via deep learning
bioRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Авг. 8, 2024
ABSTRACT
Disease
is
one
of
the
primary
factors
affecting
life
activities,
with
complex
etiologies
often
influenced
by
gene
expression
and
mutation.
Currently,
wet-lab
experiments
have
analyzed
mechanisms
mutations,
but
these
are
usually
limited
costs
wet
constraints
in
sample
types
scales.
Therefore,
this
paper
constructs
a
real-world
mutation-induced
disease
dataset
proposes
Capsule
networks
Graph
topology
multi-head
attention
(CGM)
to
predict
mutation-disease
associations.
CGM
can
accurately
protein
associations,
order
further
elucidate
pathogenicity
we
also
verified
that
mutations
lead
structural
alterations
Swiss-model,
which
suggests
conformational
changes
may
be
an
important
pathogenic
factor.
Limited
size
mutated
dataset,
performed
on
benchmark
imbalanced
datasets,
where
mined
22
unknown
interaction
pairs
from
better
illustrating
potential
predicting
In
summary,
curates
real
mutations-disease
providing
novel
tool
for
understanding
biomolecular
pathways
mechanisms.
Язык: Английский
BioStructNet: Structure-Based Network with Transfer Learning for Predicting Biocatalyst Functions
Journal of Chemical Theory and Computation,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 20, 2024
Enzyme–substrate
interactions
are
essential
to
both
biological
processes
and
industrial
applications.
Advanced
machine
learning
techniques
have
significantly
accelerated
biocatalysis
research,
revolutionizing
the
prediction
of
biocatalytic
activities
facilitating
discovery
novel
biocatalysts.
However,
limited
availability
data
for
specific
enzyme
functions,
such
as
conversion
efficiency
stereoselectivity,
presents
challenges
accuracy.
In
this
study,
we
developed
BioStructNet,
a
structure-based
deep
network
that
integrates
protein
ligand
structural
capture
complexity
enzyme–substrate
interactions.
Benchmarking
studies
with
different
algorithms
showed
enhanced
predictive
accuracy
BioStructNet.
To
further
optimize
small
set,
implemented
transfer
in
framework,
training
source
model
on
large
set
fine-tuning
it
small,
function-specific
using
CalB
case
study.
The
performance
was
validated
by
comparing
attention
heat
maps
generated
BioStructNet
interaction
module
revealed
from
molecular
dynamics
simulations
complexes.
would
accelerate
functional
enzymes
use,
particularly
cases
where
sets
small.
Язык: Английский
BioStructNet: Structure-Based Network with Transfer Learning for Predicting Biocatalyst Functions
bioRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Окт. 18, 2024
Abstract
Enzyme-substrate
interactions
are
essential
to
both
biological
processes
and
industrial
applications.
Advanced
machine
learning
techniques
have
significantly
accelerated
biocatalysis
research,
revolutionizing
the
prediction
of
biocatalytic
activities
facilitating
discovery
novel
biocatalysts.
However,
limited
availability
data
for
specific
enzyme
functions,
such
as
conversion
efficiency
stereoselectivity,
presents
challenges
accuracy.
In
this
study,
we
developed
BioStructNet,
a
structure-based
deep
network
that
integrates
protein
ligand
structural
capture
complexity
enzyme-substrate
interactions.
Benchmarking
studies
with
different
algorithms
showed
enhanced
predictive
accuracy
BioStructNet.
To
further
optimize
small
dataset,
implemented
transfer
in
framework,
training
source
model
on
large
dataset
fine-tuning
it
small,
function-specific
using
CalB
case
study.
The
performance
was
validated
by
comparing
attention
heat
maps
generated
BioStructNet
interaction
module,
substrate
revealed
complexes
from
molecular
simulations.
would
accelerate
functional
enzymes
use,
particularly
cases
where
datasets
small.
Язык: Английский
Learning allosteric interactions in Gα proteins from molecular dynamics simulations
bioRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Ноя. 1, 2024
Abstract
Gα
is
a
key
subunit
of
heterotrimeric
guanine-nucleotide-binding
regulatory
proteins,
yet
its
conformational
dynamics
are
not
fully
understood.
In
this
study,
we
developed
Transformer-based
graph
neural
network
framework,
Dynamic-Mixed
Transformer
(DMFormer),
to
investigate
Gαo.
DMFormer
achieved
an
AUC
0.75
on
the
training
set,
demonstrating
robustness
in
distinguishing
active
and
inactive
states.
The
interpretability
model
was
enhanced
using
integrated
gradients,
identifying
Switch
II
as
critical
motif
stabilizing
state
revealing
distinct
movement
patterns
between
GTPase
α-Helix
domains.
Our
findings
suggest
that
rigidity
Q205L
mutant
segment
leads
persistent
activation.
Overall,
our
study
showcases
effective
tool
for
analyzing
protein
dynamics,
offering
valuable
insights
into
activation
mechanisms
protein.
Язык: Английский