Artificial
intelligence
(AI)
models
have
been
used
to
study
the
compositional
regularities
of
proteins
in
nature,
enabling
it
assist
protein
design
improve
efficiency
engineering
and
reduce
manufacturing
cost.
However,
industrial
settings,
are
often
required
work
extreme
environments
where
they
relatively
scarce
or
even
non-existent
nature.
Since
such
almost
absent
training
datasets,
is
uncertain
whether
AI
model
possesses
capability
evolving
adapt
conditions.
Antibodies
crucial
components
affinity
chromatography,
hoped
remain
active
at
most
cannot
tolerate.
In
this
study,
we
applied
an
advanced
large
language
(LLM),
Pro-PRIME
model,
alkali
resistance
a
representative
antibody,
VHH
antibody
capable
binding
growth
hormone.
Through
two
rounds
design,
ensured
that
selected
mutant
has
enhanced
functionality,
including
higher
thermal
stability,
pH
stronger
affinity,
thereby
validating
generalized
LLM
meeting
specific
demands.
To
best
our
knowledge,
first
LLM-designed
product,
which
successfully
mass
production.
mLife,
Год журнала:
2024,
Номер
3(4), С. 492 - 504
Опубликована: Дек. 1, 2024
Optimizing
enzyme
thermostability
is
essential
for
advancements
in
protein
science
and
industrial
applications.
Currently,
(semi-)rational
design
random
mutagenesis
methods
can
accurately
identify
single-point
mutations
that
enhance
thermostability.
However,
complex
epistatic
interactions
often
arise
when
multiple
mutation
sites
are
combined,
leading
to
the
complete
inactivation
of
combinatorial
mutants.
As
a
result,
constructing
an
optimized
requires
repeated
rounds
incrementally
incorporate
single
sites,
which
highly
time-consuming.
In
this
study,
we
developed
AI-aided
strategy
engineering
efficiently
facilitates
recombination
beneficial
mutations.
We
utilized
data
from
creatinase,
including
18
mutants,
22
double-point
21
triple-point
12
quadruple-point
Using
these
as
inputs,
used
temperature-guided
language
model,
Pro-PRIME,
learn
features
After
two
design,
obtained
50
mutants
with
superior
thermostability,
achieving
success
rate
100%.
The
best
mutant,
13M4,
contained
13
maintained
nearly
full
catalytic
activity
compared
wild-type.
It
showed
10.19°C
increase
melting
temperature
~655-fold
half-life
at
58°C.
Additionally,
model
successfully
captured
epistasis
high-order
sign
(K351E)
synergistic
(D17V/I149V).
elucidated
mechanism
long-range
detail
using
dynamics
cross-correlation
matrix
method.
Our
work
provides
efficient
framework
designing
studying
effects
protein-directed
evolution.
Research Square (Research Square),
Год журнала:
2025,
Номер
unknown
Опубликована: Янв. 7, 2025
Abstract
Engineering
design,
a
cornerstone
of
technological
innovation,
faces
persistent
challenges
from
the
rigidity
traditional
methods
and
insufficient
responsiveness
emerging
AI
tools
to
fully
address
its
inherently
complex,
dynamic,
creativity-driven
demands.
Here
we
introduce
iDesignGPT,
novel
framework
that
integrates
large
language
model
with
established
design
methodologies
enable
dynamic
multi-agent
collaboration
for
problem
refinement,
information
gathering,
space
exploration,
iterative
optimization.
By
incorporating
metrics
such
as
coverage,
diversity,
novelty,
iDesignGPT
provides
decision-enabling,
data-driven
insights
conceptual
engineering
evaluation.
Our
results
reveal
surpasses
benchmark
models
in
generating
innovative,
modular,
rational
solutions,
particularly
exploratory,
open-ended
scenarios
prioritizing
creativity
adaptability.
User
studies,
involving
both
students
experienced
engineers,
validate
ability
uncover
hidden
requirements,
foster
creativity,
enhance
workflow
transparency.
Collectively,
these
findings
position
scalable
platform
lowers
expertise
barrier,
fosters
interdisciplinary
collaboration,
expands
transformative
potential
AI-assisted
design.
Artificial
intelligence
(AI)
models
have
been
used
to
study
the
compositional
regularities
of
proteins
in
nature,
enabling
it
assist
protein
design
improve
efficiency
engineering
and
reduce
manufacturing
cost.
However,
industrial
settings,
are
often
required
work
extreme
environments
where
they
relatively
scarce
or
even
non-existent
nature.
Since
such
almost
absent
training
datasets,
is
uncertain
whether
AI
model
possesses
capability
evolving
adapt
conditions.
Antibodies
crucial
components
affinity
chromatography,
hoped
remain
active
at
most
cannot
tolerate.
In
this
study,
we
applied
an
advanced
large
language
(LLM),
Pro-PRIME
model,
alkali
resistance
a
representative
antibody,
VHH
antibody
capable
binding
growth
hormone.
Through
two
rounds
design,
ensured
that
selected
mutant
has
enhanced
functionality,
including
higher
thermal
stability,
pH
stronger
affinity,
thereby
validating
generalized
LLM
meeting
specific
demands.
To
best
our
knowledge,
first
LLM-designed
product,
which
successfully
mass
production.
Artificial
intelligence
(AI)
models
have
been
used
to
study
the
compositional
regularities
of
proteins
in
nature,
enabling
it
assist
protein
design
improve
efficiency
engineering
and
reduce
manufacturing
cost.
However,
industrial
settings,
are
often
required
work
extreme
environments
where
they
relatively
scarce
or
even
non-existent
nature.
Since
such
almost
absent
training
datasets,
is
uncertain
whether
AI
model
possesses
capability
evolving
adapt
conditions.
Antibodies
crucial
components
affinity
chromatography,
hoped
remain
active
at
most
cannot
tolerate.
In
this
study,
we
applied
an
advanced
large
language
(LLM),
Pro-PRIME
model,
alkali
resistance
a
representative
antibody,
VHH
antibody
capable
binding
growth
hormone.
Through
two
rounds
design,
ensured
that
selected
mutant
has
enhanced
functionality,
including
higher
thermal
stability,
pH
resistance,
stronger
affinity,
thereby
validating
generalized
LLM
meeting
specific
demands.
To
best
our
knowledge,
first
LLM-designed
product,
which
successfully
mass
production.
Physical Review Research,
Год журнала:
2025,
Номер
7(1)
Опубликована: Фев. 28, 2025
Predicting
the
fitness
of
viral
proteins
is
fundamental
to
understanding
evolution
and
developing
antiviral
strategies.
This
study
introduces
Venus-EEM,
an
entropy-driven
ensemble
model,
aimed
at
improving
performance
zero-shot
predictions
for
protein
across
diverse
datasets.
We
demonstrate
that
entropy
serves
as
effective
criterion
selecting
optimal
models,
enabling
adaptive
model
selection
different
prediction
tasks.
By
incorporating
entropy-weighted
learning
from
multiple
language
Venus-EEM
achieves
superior
compared
existing
methods.
validate
model's
effectiveness
through
comprehensive
evaluation
on
datasets
a
detailed
case
T7
RNA
polymerase
(T7
RNAP)
activity.
Our
findings
provide
approach
predicting
mutations
based
entropy,
bridging
physics
principles
with
practical
biological
challenges.
Published
by
American
Physical
Society
2025
mLife,
Год журнала:
2025,
Номер
4(2), С. 107 - 125
Опубликована: Март 28, 2025
Abstract
Biosynthesis—a
process
utilizing
biological
systems
to
synthesize
chemical
compounds—has
emerged
as
a
revolutionary
solution
21st‐century
challenges
due
its
environmental
sustainability,
scalability,
and
high
stereoselectivity
regioselectivity.
Recent
advancements
in
artificial
intelligence
(AI)
are
accelerating
biosynthesis
by
enabling
intelligent
design,
construction,
optimization
of
enzymatic
reactions
systems.
We
first
introduce
the
molecular
retrosynthesis
route
planning
biochemical
pathway
including
single‐step
algorithms
AI‐based
design
tools.
highlight
advantages
large
language
models
addressing
sparsity
data.
Furthermore,
we
review
enzyme
discovery
methods
based
on
sequence
structure
alignment
techniques.
Breakthroughs
structural
prediction
expected
significantly
improve
accuracy
discovery.
also
summarize
for
de
novo
generation
nonnatural
or
orphan
reactions,
focusing
functional
annotation
techniques
reaction
small
molecule
similarity.
Turning
engineering,
discuss
strategies
thermostability,
solubility,
activity,
well
applications
AI
these
fields.
The
shift
from
traditional
experiment‐driven
data‐driven
computationally
driven
is
already
underway.
Finally,
present
potential
provide
perspective
future
research
directions.
envision
expanded
biocatalysis
drug
development,
green
chemistry,
complex
synthesis.
bioRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2025,
Номер
unknown
Опубликована: Май 3, 2025
Abstract
Thermostability
is
a
critical
goal
in
protein
engineering
for
applications
of
biocatalysts
and
biomedicines.
Despite
striking
advances
biomolecular
predictive
modeling,
reliably
identifying
stabilizing
mutations
remains
challenging.
Previously,
molecular
dynamics
(MD)
simulations
visual
inspection
have
been
used
as
secondary
filter
to
improve
the
success
rate
pre-selected
by
thermostability
algorithms.
However,
this
approach
suffers
from
low
throughput
subjectivity.
Here,
we
introduce
BoostMut
(Biophysical
Overview
Optimal
Stabilizing
Mutations),
computational
tool
that
standardizes
automates
mutation
filtering
analyzing
dynamic
structural
features
MD.
formalizes
principles
guiding
manual
verification,
providing
consistent
reproducible
stability
assessment.
Rigorous
benchmarking
across
multiple
datasets
showed
integrating
BoostMut’s
biophysical
analysis
improves
prediction
regardless
initial
predictor.
Given
modest
amount
existing
mutant
data,
performance
can
be
further
enhanced
with
lightweight
machine
learning
model.
Upon
experimentally
validating
predictions
on
enzyme
limonene-epoxide
hydrolase,
identified
previously
overlooked
inspection,
achieved
higher
overall
rate.
We
foresee
being
filtering,
an
integrated
step
workflows,
labelling
data
train
future
predictors.