Chemical Science,
Год журнала:
2024,
Номер
unknown
Опубликована: Ноя. 22, 2024
The
current
generation
of
large
language
models
(LLMs)
has
limited
chemical
knowledge.
Recently,
it
been
shown
that
these
LLMs
can
learn
and
predict
properties
through
fine-tuning.
Using
natural
to
train
machine
learning
opens
doors
a
wider
audience,
as
field-specific
featurization
techniques
be
omitted.
In
this
work,
we
explore
the
potential
limitations
approach.
We
studied
performance
fine-tuning
three
open-source
(GPT-J-6B,
Llama-3.1-8B,
Mistral-7B)
for
range
different
questions.
benchmark
their
performances
against
"traditional"
find
that,
in
most
cases,
approach
is
superior
simple
classification
problem.
Depending
on
size
dataset
type
questions,
also
successfully
address
more
sophisticated
problems.
important
conclusions
work
are
all
datasets
considered,
conversion
into
an
LLM
training
set
straightforward
with
even
relatively
small
leads
predictive
models.
These
results
suggest
systematic
use
guide
experiments
simulations
will
powerful
technique
any
research
study,
significantly
reducing
unnecessary
or
computations.
Green Energy & Environment,
Год журнала:
2022,
Номер
9(1), С. 54 - 70
Опубликована: Дек. 7, 2022
Membrane
technologies
are
becoming
increasingly
versatile
and
helpful
today
for
sustainable
development.
Machine
Learning
(ML),
an
essential
branch
of
artificial
intelligence
(AI),
has
substantially
impacted
the
research
development
norm
new
materials
energy
environment.
This
review
provides
overview
perspectives
on
ML
methodologies
their
applications
in
membrane
design
discovery.
A
brief
is
first
provided
with
current
bottlenecks
potential
solutions.
Through
applications-based
perspective
AI-aided
discovery,
we
further
show
how
strategies
applied
to
discovery
cycle
(including
material
design,
application,
process
knowledge
extraction),
various
systems,
ranging
from
gas,
liquid,
fuel
cell
separation
membranes.
Furthermore,
best
practices
integrating
methods
specific
application
targets
presented
ideal
paradigm
proposed.
The
challenges
be
addressed
prospects
AI
also
highlighted
end.
Abstract
For
gas
separation
and
catalysis
by
metal‐organic
frameworks
(MOFs),
diffusion
has
a
substantial
impact
on
the
process'
overall
rate,
so
it
is
necessary
to
determine
molecular
behavior
within
MOFs.
In
this
study,
an
interpretable
machine
learing
(ML)
model,
light
gradient
boosting
(LGBM),
trained
predict
diffusivity
selectivity
of
9
gases
(Kr,
Xe,
CH
4
,
N
2
H
S,
O
CO
He).
these
gases,
LGBM
displays
high
accuracy
(average
R
=
0.962)
superior
extrapolation
for
C
6
.
And
model
calculation
five
orders
magnitude
faster
than
dynamics
(MD)
simulations.
Subsequently,
using
interactive
desktop
application
developed
that
can
help
researchers
quickly
accurately
calculate
molecules
in
porous
crystal
materials.
Finally,
authors
find
difference
polarizability
(
ΔPol
)
key
factor
governing
combining
with
Shapley
additive
explanation
(SHAP).
By
ML,
optimal
MOFs
are
selected
separating
binary
mixtures
methanation.
This
work
provides
new
direction
exploring
structure‐property
relationships
realizing
rapid
diffusivity.
Journal of Materials Chemistry A,
Год журнала:
2023,
Номер
11(29), С. 15600 - 15634
Опубликована: Янв. 1, 2023
The
Troger's
base
(TB)
polymer
has
been
considered
as
promising
CO
2
separation
membrane
materials
and
have
intensively
studied.
In
the
current
work,
progress
of
TB
polymeric
membranes
for
is
summarized
analyzed.
ACS Applied Materials & Interfaces,
Год журнала:
2023,
Номер
15(13), С. 17421 - 17431
Опубликована: Март 27, 2023
Considering
the
existence
of
a
large
number
and
variety
metal-organic
frameworks
(MOFs)
ionic
liquids
(ILs),
assessing
gas
separation
potential
all
possible
IL/MOF
composites
by
purely
experimental
methods
is
not
practical.
In
this
work,
we
combined
molecular
simulations
machine
learning
(ML)
algorithms
to
computationally
design
an
composite.
Molecular
were
first
performed
screen
approximately
1000
different
1-n-butyl-3-methylimidazolium
tetrafluoroborate
([BMIM][BF4])
with
MOFs
for
CO2
N2
adsorption.
The
results
used
develop
ML
models
that
can
accurately
predict
adsorption
performances
[BMIM][BF4]/MOF
composites.
most
important
features
affect
CO2/N2
selectivity
extracted
from
utilized
generate
composite,
[BMIM][BF4]/UiO-66,
which
was
present
in
original
material
data
set.
This
composite
finally
synthesized,
characterized,
tested
separation.
Experimentally
measured
[BMIM][BF4]/UiO-66
matched
well
predicted
model,
it
found
be
comparable,
if
higher
than
previously
synthesized
reported
literature.
Our
proposed
approach
combining
will
highly
useful
any
within
seconds
compared
extensive
time
effort
requirements
methods.
Materials Today Energy,
Год журнала:
2023,
Номер
38, С. 101426 - 101426
Опубликована: Сен. 23, 2023
Hydrogen
(H2)
is
a
promising
energy
carrier
for
achieving
net
zero
carbon
emissions.
Metal
organic
frameworks
(MOFs)
and
covalent
(COFs)
have
emerged
as
strong
alternatives
to
traditional
porous
materials
highly
efficient
H2
storage
purification
applications.
With
the
very
rapid
continuous
increase
in
number
variety
of
MOFs
COFs,
early
studies
this
field
focused
on
experimental
testing
few
types
randomly
selected
recently
evolved
into
combining
computational
screening
large
material
databases
with
machine
learning
(ML).
In
review,
we
highlighted
recent
trends
merging
molecular
modeling
ML
COFs
purification.
After
reviewing
high-throughput
aiming
determine
best
candidates
adsorption
separation,
discussed
that
use
extracting
hidden
structure-performance
relations
from
simulation
results
provide
new
guidelines
inverse
design
novel
MOFs.
Finally,
addressed
current
opportunities
challenges
fusing
data
science
speed
development
innovative
adsorbent
membrane
respectively.
Journal of Membrane Science,
Год журнала:
2024,
Номер
713, С. 123256 - 123256
Опубликована: Сен. 3, 2024
Machine
learning
(ML)
has
been
rapidly
transforming
the
landscape
of
natural
sciences
and
potential
to
revolutionize
process
data
analysis
hypothesis
formulation
as
well
expand
scientific
knowledge.
ML
particularly
instrumental
in
advancement
cheminformatics
materials
science,
including
membrane
technology.
In
this
review,
we
analyze
current
state-of-the-art
membrane-related
applications
from
perspectives.
We
first
discuss
foundations
different
algorithms
design
choices.
Then,
traditional
deep
methods,
application
examples
literature,
are
reported.
also
importance
both
molecular
membrane-system
featurization.
Moreover,
follow
up
on
discussion
with
science
detail
literature
using
data-driven
methods
property
prediction
fabrication.
Various
fields
discussed,
such
reverse
osmosis,
gas
separation,
nanofiltration.
differentiate
between
downstream
predictive
tasks
generative
design.
Additionally,
formulate
best
practices
minimum
requirements
for
reporting
reproducible
studies
field
membranes.
This
is
systematic
comprehensive
review
science.
