Vol.01 No.01 (2025) 47-56

https://ojs.polmed.ac.id/index.php/SiEST

Literature Review: Computational Methods for Designing
Thermostable Efficient and Cost-Effective Enzymes for
Industrial Applications
Prisca Caesa Moneteringtyas1*, Nahzim Rahmat1, Inten Pangestika1, Sri Rahayu Widya
Ningrum2, Annisa Fillaeli3, Aliyah Aliyah4

1 Industrial Chemical Engineering Technology, Department of Mechanical Engineering, Politeknik Negeri Medan, 20155,

Medan, Indonesia

2 Study Program of Industrial Chemical Engineering, Politeknik Negeri Lampung, 35144, , Bandar Lampung, Indonesia
3 Department of Chemistry Education, Faculty of Mathematics and Natural Sciences, Universitas Negeri Yogyakarta,

Indonesia, Jl. Colombo 1, Yogyakarta 55281, Indonesia

4 Department of Chemical Science and Engineering, Graduate School of Engineering, Kobe Univesity, 1-1 Rokkodai-cho,

Nada-Ku, Kobe, Hyogo, 657-8501,Japan
*Corresponding Author: priscacaesa@polmed.ac.id

Graphical Abstract

Highlights
•
•
•
•

This review explores recent computational methods for enhancing enzyme thermostability for industrial
applications.
Structure-based design and machine learning are key approaches, each with unique strengths and
limitations.
Hybrid models integrating these techniques show improved predictive accuracy and enzyme performance.
A comparison of leading software tools is provided to guide method selection based on research goals.

ARTICLE INFO

ABSTRACT

Keywords:
Computational
methods,
enzyme, thermostability

Enzymes play a vital role as biocatalysts in various industrial
applications due to their high specificity and efficiency under
mild conditions. However, their limited thermostability
significantly constrains their operational lifespan and
effectiveness at elevated temperatures. This review examines
recent advancements in computational methods aimed at
enhancing enzyme thermostability, focusing on structure-based
rational design, machine learning, and hybrid approaches. Key
findings highlight the effectiveness of structure-based methods,
in optimizing enzyme structures, while machine learning
approaches demonstrated potential in predicting stabilizing
mutations. This review identifies key research gaps and proposes
directions for future studies to facilitate the industrial adoption of
thermostable enzymes.

Article history:
Received May 26, 2025
Revised June 19, 2025
Accepted July 10, 2025
Available online July 19,2025
https://doi.orgxxxxxxxxxxx
xxxxxxx

47

Moneteringtyas et al.

1.

Sustainable in Energy and Science Technology (2025) 47–56

Introduction

Enzymes are key biocatalysts in industry because of their high specificity as well
as efficiency [1]. However, their use in practice is frequently restricted. This is because
they lack sufficient thermal stability, which makes the enzyme not effective enough
when temperatures are elevated [2]. Improving upon the thermal stability for enzymes,
without any sacrifice of their catalytic efficiency, has therefore become a focus inside
enzyme engineering.
Current improvements in advanced computing methods present truly hopeful
answers to these specific difficulties. To predict and add mutations that make enzymes
more thermally stable, was created structure-based rational design, machine learning,
and hybrid methods [3].
The purpose of this review is to carefully review the current computational
strategies, and to further assess their effectiveness, for improvements regarding the
thermal stability of enzymes. A thorough review of nearly all of these approaches should
offer some guidance on the directions in which further studies can go in order to help
thermostable enzymes find use inside of industry.

2.

Methods

This literature review illustrated the potential role that computers could play in
improving the industrial stability of enzymes. The review reviewed journal articles
published between 2015 on-wards for the reason that those articles are considered the
latest breakthrough in developing computational tools and methods. Some journals in
2005 was also included due to its foundational contributions to structure-based rational
design.
The study evaluated the literature focused on computer tools, namely, Rosetta,
GROMACS, PyMOL, TensorFlow, Scikit-learn. Review papers and works that relied
only on laboratory studies were not included. Terms such as "thermostable enzymes
computational design" and "machine learning for stability in enzyme" were used to carry
out the search.
Searches using database sites, such as ScienceDirect, Springer, Wiley Online
Library, and IEEE Xplore, were carried out to retrieve the articles. The search was able
to cover a wide range of computer techniques, which were all practically applied in
industries.
Thus, it was possible to analyze the collected data through computational methods
and main findings derived from the research gaps studies. A particular emphasis was
therefore placed on studies that employed various computer techniques since this
method is very significant in industrial applicances.

48

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

This method also allowed a comprehensive overview of current advancements
and challenges in computational enzyme
design, as it balanced the views on existing
https://ojs.polmed.ac.id/index.php/SiEST
methods' potential and limitations.

3.

Results and Discussion

There has been significant progress in computational methods to improve enzyme
thermostability, especially through the use of structure-based rational design, machine
learning, and hybrid approaches. Each method has its own specificities as outlined in
Table 1.
Table 1 Review
Authors and
Year
Korkegian et
al. (2005) [4]

Research
Topic
Enhancing
enzyme
thermostabil
ity

Method/
Software
Rosetta
(structurebased
design)

Protein
Thermostability
Engineering

Modarres et
al., (2016) [5]

Principles of
protein
engineering
for
thermostabil
ity

PyMOL,
FoldX
(rational
design and
directed
evolution)

Successfully
improved
enzyme
thermostabili
ty through
structural
modification
s.
Identified
key factors
influencing
protein
thermostabili
ty.

Thermostability
Engineering of
Industrial
Enzymes
through
Structure-Based
Rational Design

Nezhad et
al. (2022) [6]

Rational
design for
industrial
enzyme
thermostabil
ity

Rosetta,
GROMACS
(structurebased
rational
design)

Enhanced
stability
without
compromisin
g
catalytic
efficiency.

Data-Driven
Strategies for the
Computational
Design
of
Enzyme
Thermal
Stability

Dou et al.
(2023) [7]

Data-driven
approaches
for enzyme
design

Scikit-learn,
TensorFlow
(machine
learning and
data mining)

Demonstrate
d
effectiveness
of
datadriven
strategies in
predicting
stability.

Need
for
hybrid
approaches
combining
data-driven
and rational
design.

Rational-Design
Engineering to
Improve
Enzyme
Thermostability

Pongsupasa
et al., (2023)
[8]

Improving
enzyme
stability
through

PyRosetta,
GROMACS
(sitedirected
mutagenesis

Achieved
enhanced
stability by
modifying
specific

Insufficient
analysis
of
long-term
stability
in

Title
Computational
Thermostabilizat
ion
of
an
Enzyme

49

Key Findings

Research
Gaps
Limited
exploration
of machine
learning
approaches.

Lack
of
integration
between
computation
al
and
experimental
approaches.
Limited
exploration
of
datadriven
methods.

Moneteringtyas et al.
Title

Sustainable in Energy and Science Technology (2025) 47–56
Authors and
Year

Research
Topic
rational
design

Achieving
Thermostability
of a Phytase with
Resistance up to
100 °C

Tu et al.
(2024) [9]

Enhancing
phytase
stability

Computational
Advances
in
Protein
Engineering and
Enzyme Design

Derat
&
Kamerlin.
(2022) [10]

Computatio
nal methods
in enzyme
engineering

Current
Advances
in
Design
and
Engineering
Strategies
of
Industrial
Enzymes

Dinmukham
ed et al.
(2021) [11]

Strategies
for
industrial
enzyme
design

Computational
design of an
efficient
and
thermostable
esterase
for
polylactic acid
depolymerizatio
n
A
general
TemperatureGuided
Language model
to
engineer
enhanced
Stability
and
Activity
in
Proteins

Xie et al.
(2024) [12]

Enzyme
design for
plastic
degradation

Jiang et al.,
(2024) [13]

AI-based
models for
protein
engineering

Method/
Software
and
molecular
dynamics
simulations)
Schrödinger
Suite,
NAMD
(directed
evolution
and rational
design)
AlphaFold,
PyMOL,
MODELLE
R (AI-based
tools
and
molecular
simulations)
Rosetta,
Scikit-learn
(hybrid
approaches:
rational
design
+
machine
learning)
Rosetta,
GROMACS
, AutoDock
(computatio
nal protein
design and
molecular
dynamics)
PyTorch,
Transforme
rs Library
(temperatur
e-guided
language
models)

Key Findings
amino
sites.

acid

Research
Gaps
industrial
conditions.

Successfully
stabilized
phytase
at
high
temperatures
.

Limited
application
to
other
types
of
enzymes.

Showcased
the role of AI
in
accelerating
enzyme
design.

Insufficient
focus on costeffectiveness
for industrial
applications.

Highlighted
benefits
of
combining
multiple
approaches.

Lack
of
detailed
analysis on
implementati
on
challenges.

Enhanced
degradation
efficiency
and stability
of esterase.

Limited
scalability
studies
for
industrial
applications.

Improved
both stability
and activity
of proteins
using AI.

Lack
of
comparative
studies with
traditional
methods.

Information on studies regarding computational methods for improving enzyme
thermostability is furthersummarized in Table 1. It contains the title, names of authors,
date of publication, methodology or software used, important findings, and identified
research gaps. The outcome suggests that hybrid strategies combining structure-based

50

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

design and machine learning would counterbalance the limitations of independent
approaches and would be a strong strategy
to improve enzyme thermostability.
https://ojs.polmed.ac.id/index.php/SiEST
With vivid differences with respect to efficacy and feasibility, almost all the
reviewed studies in Table 1 have successfully conducted enzyme thermal
stabilization.For example, in a work done by Tu et al. [9], extraordinary enhancement of
phytase stability was reported at 100 °C, using an approach involving a combination of
Schrödinger Suite with NAMD. However, the limitations of this method apply to a few
groups of enzymes, whereas others have never been validated in industry yet.
Furthermore, Derat & Kamerlin [10] recount the importance of AI in fast-tracking
enzyme design with software such as AlphaFold and PyMOL. Their results show AI
much fast-forwards enzyme structure prediction and fruitful alterations.
Different studies show that indeed almost all the computational techniques are
successfully applied to enhance enzyme thermostability, but those differ on their
efficiency, scalability, and applicability in industry. For instance, Tu et al. [9] reported
outstanding enhanced stability of phytase using Schrödinger Suite and NAMD, which
supports the promise of combining advanced force fields with high parallelization for
enzyme design. However, such approaches would be questionable when considering
their industrial applications due to high computation costs and the need for extensive
experimental verifications.
Similarly, the application of AI software including AlphaFold has taken to task the
reliability of predictions in protein structure as stated by Derat & Kamerlin [10].
Nevertheless, the major drawback remains high computational expenses along with
hardly any attention to cost-effectiveness, making these processes untenable for
industrial application. Solving such problems implies further resource-poor AI models
and integrating cost evaluation frameworks into computational protocols.

3. 1 New Methods to Enhance Enzyme Heat Stability
Software such as Rosetta and GROMACS are very useful in sharpening enzyme
temperature forays with specific techniques. They assist in the design of enzyme
structures by proposing alterations that stabilize the molecule, such as enhancing
hydrogen bonds and salt bridges. In 2005, Korkegian et al. [4] showed that Rosetta could
be used to modify three specific sections of an enzyme, this is also supported by [14]
which also uses Rosetta to enzyme design and activity enhancement so that it makes
enzyme was more stable and efficient when heated.
Machine learning is also playing an important role in making enzymes more
temperature-resistant. Though less widely adopted to date, it has immense potential.
Machine learning algorithms can also help researchers predict the chemical changes that
make enzymes more stable. TensorFlow and Scikit-learn also assist with creating models
that efficiently process complex data with picking up relations between system changes
in the enzyme sequence and stability. In 2023, Dou et al. [7] successfully showed that

51

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

these models are capable of predicting enzyme stability with few experiments. The
trouble is, these models can be difficult to interpret, particularly in understanding how
the changes improve the stability of enzymes. Explainable AI models can be useful in
this case since they are transparent when they point out why specific changes act in the
intended way.

3. 2 Alternatives Sofware and Their Roles
Part of running successful computational research is the decision on what
software to use. Table 2 presents a comparison of key software alternatives for molecular
dynamics simulations, machine learning, protein structure modeling, molecular
visualization, and protein programming, emphasizing their unique strengths and
suitable contexts for application.
Table 2. Software Alternatives for Designing Thermostable, Efficient, and CostEffective Enzymes

Molecular
Dynamics
Simulation

Software &
Alternatives
GROMACS
↔
NAMD
↔
AMBER

Machine
Learning
(ML)

Scikit-learn
↔
TensorFlow ↔
PyTorch ↔ Weka

Protein
Structure
Modeling

MODELLER ↔
Rosetta
↔
AlphaFold
↔
AutoDock Vina

Molecular
Visualization

PyMOL ↔ VMD

Category

Protein
Programmin
g

PyRosetta
Rosetta

↔

Protein
Design

MODELLER
Swiss-Model

↔

Detailed Explanation
GROMACS excels in speed and is open-source, suitable for
small to medium-scale simulations; NAMD is superior for
large-scale simulations due to high parallelization;
AMBER offers higher accuracy for complex proteins [15].
Scikit-learn is ideal for traditional models (regression,
SVM) with simple syntax; TensorFlow is strong for largescale production with TPU support; PyTorch is flexible for
research; Weka provides a GUI for non-coding analysis
[16].
MODELLER focuses on homology modeling with
templates, Rosetta supports de novo modeling and protein
design, AlphaFold achieves the highest accuracy in
structure prediction, and AutoDock Vina specializes in
small-molecule docking [17]
PyMOL is preferred for high-quality publication images
with detailed display control; VMD is tailored for
analyzing molecular dynamics data with robust scripting
capabilities.
Rosetta offers diverse protocols for protein design and
prediction, whereas PyRosetta allows for Python scripting
to automate complex tasks with greater flexibility.
MODELLER provides more comprehensive features for
homology modeling with better parameter control, while
Swiss-Model is more user-friendly for beginners seeking
quick results without complex configurations.

Table 2 summarizes several software tools that are used for activities such as
molecular dynamics simulations, machine learning, protein structure prediction,
molecular visualization, and protein design. When selecting which software is best, it is
important to consider the research requirements, i.e., accuracy, scalability, and
computation costs. In some cases, merging and aligning different software tools will
enhance the computation performance [18].
52

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

This side-by-side comparison illustrates the advantage of selecting your software
according to the particular demands https://ojs.polmed.ac.id/index.php/SiEST
of your research environment. To illustrate, when
choosing between NAMD and GROMACS for molecular dynamics computation, you
would need to consider the size as well as the complexity of the system that is under
study. GROMACS would most suitably be applicable to small systems, while NAMD is
ideally suited for large systems due to its scalability architecture [19].

3. 3 Synergistic Potential of Hybrid Approaches
Hybrid approaches, where structure-based design and machine learning are
combined, have been suggested as a remedy to surpass the drawback of using one
approach in silico. A recent study by Dinmukhamed et al. [11] showed that hybrid
models, where structural information is utilized for training machine learning models,
performed significantly better than single approaches for predicting stabilizing
mutations. This fusion brings together the predictiveness of machine learning and the
structural precision of structure-based methods to enable possibilities of mutations to be
explored even further. Using the incorporation of Rosetta's structural modeling into
TensorFlow's machine learning framework as an example, this hybrid approach resulted
in the creation of enzymes that not only performed better but even proved more stable
in character, actually bridging the accuracy-scalability gap.

3. 4 Effectiveness and Limitations of Methods
All these computational approaches discussed here have both strengths and
weaknesses. Structure-based design is excellent at giving very high-level control over
structural modification but is very time-consuming and typically limited to small
enzyme sizes [5]. Machine learning, on the other hand, while able to process big data
very effectively, is not very transparent and requires a lot of training data [12].
The combination of rational design and machine learning offers a possible solution
to such complexities. For example, Dinmukhamed et al. [11] established that the
combination of these approaches not only enhanced the stability of enzymes but also
accelerated the designing process and ensured more accurate results. In this method,
structure simulation data are employed to train machine learning models, which then
aid in designing improved mutation designs.

3. 5 Cost-Effectiveness of Computational and Hybrid Methods in Enzyme
Design
In comparison to traditional experimental approaches, computer-aided and
hybrid designs reduce the costs by incurring lower trial-and-error testing and cheaper
reagent, equipment, and labor inputs [20]. Computer programming environments like
Rosetta, GROMACS, TensorFlow, and Scikit-learn offer predictive accuracy through
which scientists filter and rank more viable enzyme mutations to minimize expensive
experimental testing. The computer assistances accelerate enzyme design workflow in
terms of shorter development timelines and fewer overall expenses [21].
53

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

Also, these computational approaches can reduce failure rates through intentional
design and predictive computation, which equate to fewer failed experiments and their
costs. Their scalability, enabled by software and cloud computing, enable one to handle
large volumes of data without spending much more money, which makes them suitable
for industrial use. While the initial investment in computing hardware may be high, the
long-term return on investment includes reduced dependence on trial-and-error
approaches, compressed time-to-market for thermostable enzymes, and increased return
on investment (ROI).

4.

Conclusion

The review highlights the way computational approaches have compressed the
design process for thermostable enzymes, through the combination of machine learning
and structure-based rational design. The construction of hybrid strategies that blend
these methods provides a promising route for enhancing enzyme stability without loss
of catalytic function. Yet, there are a number of fundamental issues to be addressed, such
as scalability, cost, and the need for thorough experimental validation of computational
design predictions. These issues will be addressed through concerted research efforts,
the establishment of standard experimental protocols, and calibration of calculation
devices for wider industrial application. Further emphasis must be placed on the
integration of data-driven methods and experimental back-loops to iteratively refine the
design process and also to establish the practical viability of thermostable enzymes.
Having overcome systematically these challenges, we can now employ the full power of
computational tools to revolutionize enzyme engineering in industry.
Acknowledgements
The authors would like to express their sincere gratitude to the Industrial Chemical
Engineering Technology program, Department of Mechanical Engineering, Politeknik
Negeri Medan, for providing access to literature resources and continuous academic
support throughout the preparation of this review. The authors also appreciate the
valuable feedback from colleagues, which contributed to improving the clarity and
depth of this manuscript.
CRediT Authorship Contribution Statement
Prisca Caesa Moneteringtyas: Conceptualization, Supervision, Writing – review &
editing, Project administration. Nahzim Rahmat: Investigation, Formal analysis,
Visualization. Inten Pangestika: Methodology, Writing – original draft, Data curation,
Sri Rahayu Widya Ningrum: Software, Literature search, Annisa Fillaeli: Resources,
Validation, Writing – review & editing. Aliyah Aliyah:Writing – review & editing,
translating.
Conflicts of Interest
The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.

54

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

References
[1]

[2]

[3]

[4]

[5]
[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

https://ojs.polmed.ac.id/index.php/SiEST
Choi JM, Han SS, Kim HS. Industrial
applications of enzyme biocatalysis: Current
status
and
future
aspects.
Bioechnol
Adv
2015;33(7):1443-1454.
https://doi.org/10.1016/j.biotechadv.2015.02.014
Rahban M, Zolghadri S, Salehi N, Ahmad F, Haertlé T, Rezaei-Ghaleh N, et al.
Thermal stability enhancement: Fundamental concepts of protein engineering
strategies to manipulate the flexible structure. Int J Biol Macromol 2022;214:642–
654. https://doi.org/10.1016/j.ijbiomac.2022.06.154
Zheng N, Cai Y, Zhang Z, Zhou H, Deng Y, Du S, et al. Tailoring industrial
enzymes for thermostability and activity evolution by the machine learning-based
iCASE strategy. Nat Commun 2025;16:604. https://doi.org/10.1038/s41467-02555944-5
Korkegian A, Black ME, Baker D, Stoddard BL. Computational
Thermostabilization of an Enzyme. Science (1979) 2005;308:857–860.
https://doi.org/10.1126/science.1107387
Modarres HP, Mofrad MR, Sanati-Nezhad A. Protein thermostability engineering.
RSC Adv 2016;6(116):115252–115270. https://doi.org/10.1039/C6RA16992A
Nezhad NG, Rahman RNZRA, Normi YM, Oslan SN, Shariff FM, Leow TC.
Thermostability engineering of industrial enzymes through structure
modification.
Appl
Microbiol
Biotechnol
2022;106:4845–4866.
https://doi.org/10.1007/s00253-022-12067-x
Dou Z, Sun Y, Jiang X, Wu X, Li Y, Gong B, et al. Data-driven strategies for the
computational design of enzyme thermal stability: trends, perspectives, and
prospects. Acta Biochim Biophys Sin (Shanghai) 2023;55:343–355.
https://doi.org/10.3724/abbs.2023033
Pongsupasa V, Anuwan P, Maenpuen S, Wongnate T. Rational-Design
Engineering to Improve Enzyme Thermostability. In: Magnani F, Marabelli C,
Paradisi F, editors. Enzyme Engineering: Methods and Protocols, New York, NY:
Springer US; 2022, p. 159–178. https://doi.org/10.1007/978-1-0716-1826-4_9
Tu T, Wang Q, Dong R, Liu X, Penttinen L, Hakulinen N, et al. Achieving
thermostability of a phytase with resistance up to 100 oC. Journal of Biological
Chemistry 2024;300(12). https://doi.org/10.1016/j.jbc.2024.107992
Derat E, Kamerlin SCL. Computational Advances in Protein Engineering and
Enzyme Design. J Phys Chem B 2022;126(13):2449–2451.
https://doi.org/10.1021/acs.jpcb.2c01198
Dinmukhamed T, Huang Z, Liu Y, Lv X, Li J, Du G, et al. Current advances in
design and engineering strategies of industrial enzymes. Systems Microbiology
and Biomanufacturing 2021;1(1):15–23. https://doi.org/10.1007/s43393-02000005-9
Xie B, Zhang J, Sun H, Bai R, Lu D, Zhu Y, et al. Computational design of an
efficient and thermostable esterase for polylactic acid depolymerization. Green
Chemistry 2024;26(12):7268–7279. https://doi.org/10.1039/D3GC04888H
Jiang F, Li M, Dong J, Yu Y, Sun X, Wu B, et al. A general temperature-guided
language model to design proteins of enhanced stability and activity. Sci Adv
2024;10:eadr2641. https://doi.org/10.1126/sciadv.adr2641
Sun R, Wu D, Chen P, Zheng P. Cutting-edge computational approaches in
enzyme design and activity enhancement. Biochem Eng J 2024;212:109510.
https://doi.org/10.1016/j.bej.2024.109510

55

Moneteringtyas et al.

Sustainable in Energy and Science Technology (2025) 47–56

[15] Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel
RD, Kalé L, Schulten K. Scalable molecular dynamics with NAMD. J Comput
Chem. 2005 Dec;26(16):1781-1802. https://doi.org/10.1002/jcc.20289
[16] Abadi M, Barham P, Chen J, Chen Z, Davis A, Dean J, et al. TensorFlow: A system
for large-scale machine learning 2016.https://doi.org/10.48550/arXiv.1605.08695
[17] Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly
accurate protein structure prediction with AlphaFold. Nature 2021;596:583–589.
https://doi.org/10.1038/s41586-021-03819-2
[18] Krylov A, Windus TL, Barnes T, Marin-Rimoldi E, Nash JA, Pritchard B, et al.
Perspective: Computational chemistry software and its advancement as illustrated
through three grand challenge cases for molecular science. J Chem Phys
2018;149:180901. https://doi.org/10.1063/1.5052551
[19] A. Poghosyan, A. ARTASHES Shahinyan, A. H. Poghosyan, G. A. Yeghiazaryan,
H. H. Gharabekyan, and A. A. Shahinyan, “The GROMACS and NAMD Software
Packages Comparison,” Commun Comput Phys, vol. 1, no. 4, pp. 736–743, Aug. 2006,
[Online]. Available: http://www.global-sci.com/
[20] Landes SJ, McBain SA, Curran GM. An introduction to effectivenessimplementation hybrid designs. Psychiatry Res 2019;280:112513.
https://doi.org/10.1016/j.psychres.2019.112513
[21] Landwehr GM, Bogart JW, Magalhaes C, Hammarlund EG, Karim AS, Jewett MC.
Accelerated enzyme engineering by machine-learning guided cell-free expression.
Nat Commun 2025;16:865. https://doi.org/10.1038/s41467-024-55399-0

56