Unsupervised Learning of Progress Coordinates during Weighted Ensemble Simulations: Application to NTL9 Protein Folding
Journal of Chemical Theory and Computation,
Год журнала:
2025,
Номер
unknown
Опубликована: Март 19, 2025
A
major
challenge
for
many
rare-event
sampling
strategies
is
the
identification
of
progress
coordinates
that
capture
slowest
relevant
motions.
Machine-learning
methods
can
identify
in
an
unsupervised
manner
have
therefore
been
great
interest
to
simulation
community.
Here,
we
developed
a
general
method
identifying
"on-the-fly"
during
weighted
ensemble
(WE)
via
deep
learning
(DL)
outliers
among
sampled
conformations.
Our
identifies
latent
space
model
system's
conformations
periodically
trained
using
convolutional
variational
autoencoder.
As
proof
principle,
applied
our
DL-enhanced
WE
simulate
NTL9
protein
folding
process.
To
enable
rapid
tests,
simulations
propagated
discrete-state
synthetic
molecular
dynamics
trajectories
generative,
fine-grained
Markov
state
model.
Results
revealed
on-the-fly
DL
enhanced
efficiency
by
>3-fold
estimating
rate
constant.
efforts
are
significant
step
forward
slow
rare
event
sampling.
Язык: Английский
Machine Learning and Statistical Mechanics: Shared Synergies for Next Generation of Chemical Theory and Computation
Journal of Chemical Theory and Computation,
Год журнала:
2025,
Номер
unknown
Опубликована: Май 9, 2025
Язык: Английский
Graph Neural Network-State Predictive Information Bottleneck (GNN-SPIB) approach for learning molecular thermodynamics and kinetics
Digital Discovery,
Год журнала:
2024,
Номер
4(1), С. 211 - 221
Опубликована: Ноя. 28, 2024
We
present
a
graph-based
differentiable
representation
learning
method
from
atomic
coordinates
for
enhanced
sampling
methods
to
learn
both
thermodynamic
and
kinetic
properties
of
system.
Язык: Английский
Flow Matching for Optimal Reaction Coordinates of Biomolecular Systems
Journal of Chemical Theory and Computation,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 19, 2024
We
present
flow
matching
for
reaction
coordinates
(FMRC),
a
novel
deep
learning
algorithm
designed
to
identify
optimal
(RC)
in
biomolecular
reversible
dynamics.
FMRC
is
based
on
the
mathematical
principles
of
lumpability
and
decomposability,
which
we
reformulate
into
conditional
probability
framework
efficient
data-driven
optimization
using
generative
models.
While
does
not
explicitly
learn
well-established
transfer
operator
or
its
eigenfunctions,
it
can
effectively
encode
dynamics
leading
eigenfunctions
system
low-dimensional
RC
space.
further
quantitatively
compare
performance
with
several
state-of-the-art
algorithms
by
evaluating
quality
Markov
state
models
(MSM)
constructed
their
respective
spaces,
demonstrating
superiority
three
increasingly
complex
systems.
In
addition,
successfully
demonstrated
efficacy
bias
deposition
enhanced
sampling
simple
model
system.
Finally,
discuss
potential
applications
downstream
such
as
methods
MSM
construction.
Язык: Английский