Drug Design Development and Therapy,
Год журнала:
2023,
Номер
Volume 17, С. 2691 - 2725
Опубликована: Сен. 1, 2023
Abstract:
Artificial
intelligence
(AI)
and
machine
learning
(ML)
represent
significant
advancements
in
computing,
building
on
technologies
that
humanity
has
developed
over
millions
of
years—from
the
abacus
to
quantum
computers.
These
tools
have
reached
a
pivotal
moment
their
development.
In
2021
alone,
U.S.
Food
Drug
Administration
(FDA)
received
100
product
registration
submissions
heavily
relied
AI/ML
for
applications
such
as
monitoring
improving
human
performance
compiling
dossiers.
To
ensure
safe
effective
use
drug
discovery
manufacturing,
FDA
numerous
other
federal
agencies
issued
continuously
updated,
stringent
guidelines.
Intriguingly,
these
guidelines
are
often
generated
or
updated
with
aid
themselves.
The
overarching
goal
is
expedite
discovery,
enhance
safety
profiles
existing
drugs,
introduce
novel
treatment
modalities,
improve
manufacturing
compliance
robustness.
Recent
publications
offer
an
encouraging
outlook
potential
tools,
emphasizing
need
careful
deployment.
This
expanded
market
opportunities
retraining
personnel
handling
enabled
innovative
emerging
therapies
gene
editing,
CRISPR-Cas9,
CAR-T
cells,
mRNA-based
treatments,
personalized
medicine.
summary,
maturation
testament
ingenuity.
Far
from
being
autonomous
entities,
created
by
humans
designed
solve
complex
problems
now
future.
paper
aims
present
status
technologies,
along
examples
future
applications.
Keywords:
FDA,
artificial
intelligence,
learning,
development,
advanced
Nature Communications,
Год журнала:
2022,
Номер
13(1)
Опубликована: Янв. 10, 2022
Highly
accurate
protein
structure
predictions
by
deep
neural
networks
such
as
AlphaFold2
and
RoseTTAFold
have
tremendous
impact
on
structural
biology
beyond.
Here,
we
show
that,
although
these
learning
approaches
originally
been
developed
for
the
in
silico
folding
of
monomers,
also
enables
quick
modeling
peptide-protein
interactions.
Our
simple
implementation
generates
complex
models
without
requiring
multiple
sequence
alignment
information
peptide
partner,
can
handle
binding-induced
conformational
changes
receptor.
We
explore
what
has
memorized
learned,
describe
specific
examples
that
highlight
differences
compared
to
state-of-the-art
docking
protocol
PIPER-FlexPepDock.
These
results
holds
great
promise
providing
insight
into
a
wide
range
complexes,
serving
starting
point
detailed
characterization
manipulation
Nature,
Год журнала:
2023,
Номер
616(7958), С. 673 - 685
Опубликована: Апрель 26, 2023
Computer-aided
drug
discovery
has
been
around
for
decades,
although
the
past
few
years
have
seen
a
tectonic
shift
towards
embracing
computational
technologies
in
both
academia
and
pharma.
This
is
largely
defined
by
flood
of
data
on
ligand
properties
binding
to
therapeutic
targets
their
3D
structures,
abundant
computing
capacities
advent
on-demand
virtual
libraries
drug-like
small
molecules
billions.
Taking
full
advantage
these
resources
requires
fast
methods
effective
screening.
includes
structure-based
screening
gigascale
chemical
spaces,
further
facilitated
iterative
approaches.
Highly
synergistic
are
developments
deep
learning
predictions
target
activities
lieu
receptor
structure.
Here
we
review
recent
advances
technologies,
potential
reshaping
whole
process
development,
as
well
challenges
they
encounter.
We
also
discuss
how
rapid
identification
highly
diverse,
potent,
target-selective
ligands
protein
can
democratize
process,
presenting
new
opportunities
cost-effective
development
safer
more
small-molecule
treatments.
Recent
approaches
application
streamlining
discussed.
Proteins Structure Function and Bioinformatics,
Год журнала:
2021,
Номер
89(12), С. 1607 - 1617
Опубликована: Сен. 17, 2021
Critical
assessment
of
structure
prediction
(CASP)
is
a
community
experiment
to
advance
methods
computing
three-dimensional
protein
from
amino
acid
sequence.
Core
components
are
rigorous
blind
testing
and
evaluation
the
results
by
independent
assessors.
In
most
recent
(CASP14),
deep-learning
one
research
group
consistently
delivered
computed
structures
rivaling
corresponding
experimental
ones
in
accuracy.
this
sense,
represent
solution
classical
protein-folding
problem,
at
least
for
single
proteins.
The
models
have
already
been
shown
be
capable
providing
solutions
problematic
crystal
structures,
there
broad
implications
rest
structural
biology.
Other
groups
also
substantially
improved
performance.
Here,
we
describe
these
outline
some
many
implications.
related
areas
CASP,
including
modeling
complexes,
refinement,
estimation
model
accuracy,
inter-residue
contacts
distances,
described.
Equilibrium
fluctuations
and
triggered
conformational
changes
often
underlie
the
functional
cycles
of
membrane
proteins.
For
example,
transporters
mediate
passage
molecules
across
cell
membranes
by
alternating
between
inward-
outward-facing
states,
while
receptors
undergo
intracellular
structural
rearrangements
that
initiate
signaling
cascades.
Although
plasticity
these
proteins
has
historically
posed
a
challenge
for
traditional
de
novo
protein
structure
prediction
pipelines,
recent
success
AlphaFold2
(AF2)
in
CASP14
culminated
modeling
transporter
multiple
conformations
to
high
accuracy.
Given
AF2
was
designed
predict
static
structures
proteins,
it
remains
unclear
if
this
result
represents
an
underexplored
capability
accurately
and/or
heterogeneity.
Here,
we
present
approach
drive
sample
alternative
topologically
diverse
G-protein-coupled
are
absent
from
training
set.
Whereas
models
most
generated
using
default
pipeline
conformationally
homogeneous
nearly
identical
one
another,
reducing
depth
input
sequence
alignments
stochastic
subsampling
led
generation
accurate
conformations.
In
our
benchmark,
spanned
range
two
experimental
interest,
with
at
extremes
distributions
observed
be
among
(average
template
score
0.94).
These
results
suggest
straightforward
identifying
native-like
also
highlighting
need
next
deep
learning
algorithms
ensembles
biophysically
relevant
states.
Proteins Structure Function and Bioinformatics,
Год журнала:
2021,
Номер
89(12), С. 1711 - 1721
Опубликована: Окт. 4, 2021
We
describe
the
operation
and
improvement
of
AlphaFold,
system
that
was
entered
by
team
AlphaFold2
to
"human"
category
in
14th
Critical
Assessment
Protein
Structure
Prediction
(CASP14).
The
AlphaFold
CASP14
is
entirely
different
one
CASP13.
It
used
a
novel
end-to-end
deep
neural
network
trained
produce
protein
structures
from
amino
acid
sequence,
multiple
sequence
alignments,
homologous
proteins.
In
assessors'
ranking
summed
z
scores
(>2.0),
scored
244.0
compared
90.8
next
best
group.
predictions
made
had
median
domain
GDT_TS
92.4;
this
first
time
level
average
accuracy
has
been
achieved
during
CASP,
especially
on
more
difficult
Free
Modeling
targets,
represents
significant
state
art
structure
prediction.
reported
how
run
as
human
improved
such
it
now
achieves
an
equivalent
performance
without
intervention,
opening
door
highly
accurate
large-scale
High-resolution
experimental
structural
determination
of
protein-protein
interactions
has
led
to
valuable
mechanistic
insights,
yet
due
the
massive
number
and
limitations
there
is
a
need
for
computational
methods
that
can
accurately
model
their
structures.
Here
we
explore
use
recently
developed
deep
learning
method,
AlphaFold,
predict
structures
protein
complexes
from
sequence.
With
benchmark
152
diverse
heterodimeric
complexes,
multiple
implementations
parameters
AlphaFold
were
tested
accuracy.
Remarkably,
many
cases
(43%)
had
near-native
models
(medium
or
high
critical
assessment
predicted
accuracy)
generated
as
top-ranked
predictions
by
greatly
surpassing
performance
unbound
docking
(9%
success
rate
models),
however
modeling
antibody-antigen
within
our
set
was
unsuccessful.
We
identified
sequence
features
associated
with
lack
success,
also
investigated
impact
alignment
input.
Benchmarking
multimer-optimized
version
(AlphaFold-Multimer)
released
confirmed
low
(11%
success),
found
T
cell
receptor-antigen
are
likewise
not
modeled
algorithm,
showing
adaptive
immune
recognition
poses
challenge
current
algorithm
model.
Overall,
study
demonstrates
end-to-end
transient
highlights
areas
improvement
future
developments
reliably
any
interaction
interest.
INTRODUCTION
The
eukaryotic
nucleus
pro-tects
the
genome
and
is
enclosed
by
two
membranes
of
nuclear
envelope.
Nuclear
pore
complexes
(NPCs)
perforate
envelope
to
facilitate
nucleocytoplasmic
transport.
With
a
molecular
weight
∼120
MDa,
human
NPC
one
larg-est
protein
complexes.
Its
~1000
proteins
are
taken
in
multiple
copies
from
set
about
30
distinct
nucleoporins
(NUPs).
They
can
be
roughly
categorized
into
classes.
Scaf-fold
NUPs
contain
folded
domains
form
cylindrical
scaffold
architecture
around
central
channel.
Intrinsically
disordered
line
extend
channel,
where
they
interact
with
cargo
highly
dynamic.
It
responds
changes
tension
conforma-tional
breathing
that
manifests
dilation
constriction
movements.
Elucidating
architecture,
ultimately
at
atomic
resolution,
will
important
for
gaining
more
precise
understanding
function
dynamics
but
imposes
substantial
chal-lenge
structural
biologists.
RATIONALE
Considerable
progress
has
been
made
toward
this
goal
joint
effort
field.
A
synergistic
combination
complementary
approaches
turned
out
critical.
In
situ
biology
techniques
were
used
reveal
overall
layout
defines
spatial
reference
modeling.
High-resolution
structures
many
determined
vitro.
Proteomic
analysis
extensive
biochemical
work
unraveled
interaction
network
NUPs.
Integra-tive
modeling
combine
different
types
data,
resulting
rough
outline
scaffold.
Previous
struc-tural
models
NPC,
however,
patchy
limited
accuracy
owing
several
challenges:
(i)
Many
high-resolution
individual
have
solved
distantly
related
species
and,
consequently,
do
not
comprehensively
cover
their
counterparts.
(ii)
scaf-fold
interconnected
intrinsically
linker
straight-forwardly
accessible
common
techniques.
(iii)
intimately
embraces
fused
inner
outer
distinctive
topol-ogy
cannot
studied
isolation.
(iv)
conformational
limits
resolution
achievable
structure
determination.
RESULTS
study,
we
artificial
intelligence
(AI)-based
prediction
generate
an
exten-sive
repertoire
subcomplexes.
various
interfaces
so
far
remained
structurally
uncharac-terized.
Benchmarking
against
previous
unpublished
x-ray
cryo-electron
micros-copy
revealed
unprecedented
accu-racy.
We
obtained
well-resolved
tomographic
maps
both
constricted
dilated
states
hu-man
NPC.
Using
integrative
modeling,
fit-ted
microscopy
maps.
explicitly
included
traced
trajectory
through
scaf-fold.
elucidated
great
detail
how
mem-brane-associated
transmembrane
distributed
across
fusion
topology
membranes.
architectural
model
increases
coverage
twofold.
extensively
validated
our
earlier
new
experimental
data.
completeness
enabled
microsecond-long
coarse-grained
simulations
within
explicit
membrane
en-vironment
solvent.
These
prevents
otherwise
stable
double-membrane
small
diameters
absence
tension.
CONCLUSION
Our
70-MDa
atomically
re-solved
covers
>90%
captures
occur
during
constriction.
also
reveals
anchoring
sites
NUPs,
identification
which
prerequisite
complete
dy-namic
study
exempli-fies
AI-based
may
accelerate
elucidation
subcellular
ar-chitecture
resolution.
[Figure:
see
text].
Chemical Science,
Год журнала:
2023,
Номер
14(6), С. 1443 - 1452
Опубликована: Янв. 1, 2023
The
application
of
artificial
intelligence
(AI)
has
been
considered
a
revolutionary
change
in
drug
discovery
and
development.
In
2020,
the
AlphaFold
computer
program
predicted
protein
structures
for
whole
human
genome,
which
remarkable
breakthrough
both
AI
applications
structural
biology.
Despite
varying
confidence
levels,
these
could
still
significantly
contribute
to
structure-based
design
novel
targets,
especially
ones
with
no
or
limited
information.
this
work,
we
successfully
applied
our
end-to-end
AI-powered
engines,
including
biocomputational
platform
PandaOmics
generative
chemistry
Chemistry42.
A
hit
molecule
against
target
without
an
experimental
structure
was
identified,
starting
from
selection
towards
identification,
cost-
time-efficient
manner.
provided
interest
treatment
hepatocellular
carcinoma
(HCC)
Chemistry42
generated
molecules
based
on
by
AlphaFold,
selected
were
synthesized
tested
biological
assays.
Through
approach,
identified
small
compound
cyclin-dependent
kinase
20
(CDK20)
binding
constant
Kd
value
9.2
±
0.5
μM
(n
=
3)
within
30
days
after
only
synthesizing
7
compounds.
Based
available
data,
second
round
generation
conducted
through
this,
more
potent
molecule,
ISM042-2-048,
discovered
average
566.7
256.2
nM
3).
Compound
ISM042-2-048
also
showed
good
CDK20
inhibitory
activity
IC50
33.4
22.6
addition,
demonstrated
selective
anti-proliferation
HCC
cell
line
overexpression,
Huh7,
208.7
3.3
nM,
compared
counter
screen
HEK293
(IC50
1706.7
670.0
nM).
This
work
is
first
demonstration
applying
identification
process
discovery.
Proteins Structure Function and Bioinformatics,
Год журнала:
2022,
Номер
90(11), С. 1873 - 1885
Опубликована: Май 5, 2022
The
family
of
G-protein
coupled
receptors
(GPCRs)
is
one
the
largest
protein
families
in
human
genome.
GPCRs
transduct
chemical
signals
from
extracellular
to
intracellular
regions
via
a
conformational
switch
between
active
and
inactive
states
upon
ligand
binding.
While
experimental
structures
remain
limited,
high-accuracy
computational
predictions
are
now
possible
with
AlphaFold2.
However,
AlphaFold2
only
predicts
state
biased
toward
either
or
conformation
depending
on
GPCR
class.
Here,
multi-state
prediction
protocol
introduced
that
extends
predict
at
very
high
accuracy
using
state-annotated
templated
databases.
predicted
models
accurately
capture
main
structural
changes
activation
atomic
level.
For
most
benchmarked
(10
out
15),
were
closer
their
corresponding
structures.
Median
RMSDs
transmembrane
1.12
Å
1.41
for
models,
respectively.
more
suitable
protein-ligand
docking
than
original
template-based
models.
Finally,
our
accurate
GPCR-peptide
complex
Dock
2021,
blind
GPCR-ligand
modeling
competition.
We
expect
both
will
promote
understanding
mechanisms
drug
discovery
GPCRs.
At
time,
new
paves
way
towards
capturing
dynamics
proteins
machine-learning
methods.
Frontiers in Bioinformatics,
Год журнала:
2023,
Номер
3
Опубликована: Фев. 28, 2023
Three-dimensional
protein
structure
is
directly
correlated
with
its
function
and
determination
critical
to
understanding
biological
processes
addressing
human
health
life
science
problems
in
general.
Although
new
structures
are
experimentally
obtained
over
time,
there
still
a
large
difference
between
the
number
of
sequences
placed
Uniprot
those
resolved
tertiary
structure.
In
this
context,
studies
have
emerged
predict
by
methods
based
on
template
or
free
modeling.
last
years,
different
been
combined
overcome
their
individual
limitations,
until
emergence
AlphaFold2,
which
demonstrated
that
predicting
high
accuracy
at
unprecedented
scale
possible.
Despite
current
impact
field,
AlphaFold2
has
limitations.
Recently,
language
models
promised
revolutionize
structural
biology
allowing
discovery
only
from
evolutionary
patterns
present
sequence.
Even
though
these
do
not
reach
accuracy,
they
already
covered
some
being
able
more
than
200
million
proteins
metagenomic
databases.
mini-review,
we
provide
an
overview
breakthroughs
prediction
before
after
emergence.