Active learning-assisted directed evolution

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: Jan. 16, 2025

Abstract Directed evolution (DE) is a powerful tool to optimize protein fitness for specific application. However, DE can be inefficient when mutations exhibit non-additive, or epistatic, behavior. Here, we present Active Learning-assisted Evolution (ALDE), an iterative machine learning-assisted workflow that leverages uncertainty quantification explore the search space of proteins more efficiently than current methods. We apply ALDE engineering landscape challenging DE: optimization five epistatic residues in active site enzyme. In three rounds wet-lab experimentation, improve yield desired product non-native cyclopropanation reaction from 12% 93%. also perform computational simulations on existing sequence-fitness datasets support our argument effective DE. Overall, practical and broadly applicable strategy unlock improved outcomes.

Language: Английский

---

Effective Gene Expression Prediction and Optimization from Protein Sequences

Advanced Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 9, 2025

Abstract High soluble protein expression in heterologous hosts is crucial for various research and applications. Despite considerable on the impact of codon usage levels, relationship between sequence often overlooked. In this study, a novel connection uncovered, leading to development SRAB (Strength Relative Amino Acid Bias) based AEI (Amino Expression Index). The served as an objective measure correlation, with higher values enhancing expression. Subsequently, pre‐trained model MP‐TRANS (MindSpore Protein Transformer) developed fine‐tuned using transfer learning techniques create 88 prediction models (MPB‐EXP) predicting levels across species. This approach achieved average accuracy 0.78, surpassing conventional machine methods. Additionally, mutant generation model, MPB‐MUT, devised utilized enhance specific hosts. Experimental validation demonstrated that top 3 mutants xylanase (previously not expressed Escherichia coli ) successfully high‐level E. . These findings highlight efficacy optimizing gene sequences.

Language: Английский

---

Engineering highly active nuclease enzymes with machine learning and high-throughput screening

Cell Systems, Journal Year: 2025, Volume and Issue: 16(3), P. 101236 - 101236

Published: March 1, 2025

Highlights•TeleProt is a method for combining evolutionary and assay data to design novel proteins•TeleProt achieved an improved hit rate diversity compared with directed evolution•TeleProt discovered nuclease enzyme 11-fold-improved specific activity•Zero-shot showed higher relative error-prone PCRSummaryOptimizing enzymes function in chemical environments central goal of synthetic biology, but optimization often hindered by rugged fitness landscape costly experiments. In this work, we present TeleProt, machine learning (ML) framework that blends experimental diverse protein libraries, employ it improve the catalytic activity degrades biofilms accumulate on chronic wounds. After multiple rounds high-throughput experiments, TeleProt found significantly better top-performing than evolution (DE), had at finding diverse, high-activity variants, was even able high-performance initial library using no prior data. We have released dataset 55,000 one most extensive genotype-phenotype landscapes date, drive further progress ML-guided design. A record paper's transparent peer review process included supplemental information.Graphical abstract

Language: Английский

---

Fluorogenic, Subsingle-Turnover Monitoring of Enzymatic Reactions Involving NAD(P)H Provides a Generalized Platform for Directed Ultrahigh-Throughput Evolution of Biocatalysts in Microdroplets

Journal of the American Chemical Society, Journal Year: 2025, Volume and Issue: unknown

Published: March 24, 2025

Enzyme engineering and discovery are crucial for a sustainable future bioeconomy. Harvesting new biocatalysts from large libraries through directed evolution or functional metagenomics requires accessible, rapid assays. Ultrahigh-throughput screening formats often require optical readouts, leading to the use of model substrates that may misreport target activity necessitate bespoke synthesis. This is particular challenge when glycosyl hydrolases, which leverage molecular recognition beyond glycosidic bond, so complex chemical synthesis would have be deployed build fluoro- chromogenic substrate. In contrast, coupled assays represent modular "plug-and-play" system: any enzyme–substrate pairing can investigated, provided reaction produce common intermediate links catalytic detection cascade readout. Here, we establish producing fluorescent readout in response NAD(P)H via glutathione reductase subsequent thiol-mediated uncaging reaction, with low nanomolar limit plates. Further scaling down microfluidic droplet possible: fluorophore leakage-free report 3 orders magnitude-improved sensitivity compared absorbance-based systems, resolution 361,000 product molecules per droplet. Our approach enables nonfluorogenic droplet-based enrichments, applicability hydrolases imine reductases (IREDs). To demonstrate assay's readiness combinatorial experiments, one round was performed select glycosidase processing natural substrate, beechwood xylan, improved kinetic parameters pool >106 mutagenized sequences.

Language: Английский

---

Active Learning-Assisted Directed Evolution

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 28, 2024

ABSTRACT Directed evolution (DE) is a powerful tool to optimize protein fitness for specific application. However, DE can be inefficient when mutations exhibit non-additive, or epistatic, behavior. Here, we present Active Learning-assisted Evolution (ALDE), an iterative machine learning-assisted workflow that leverages uncertainty quantification explore the search space of proteins more efficiently than current methods. We apply ALDE engineering landscape challenging DE: optimization five epistatic residues in active site enzyme. In three rounds wet-lab experimentation, improve yield desired product non-native cyclopropanation reaction from 12% 93%. also perform computational simulations on existing sequence-fitness datasets support our argument effective DE. Overall, practical and broadly applicable strategy unlock improved outcomes.

Language: Английский

---

Evolutionary Specialisation of a Promiscuous Artificial Enzyme

Published: Sept. 2, 2024

The evolution of a promiscuous enzyme for its various activities often results in catalytically specialized variants. This is an important natural mechanism to ensure the proper functioning metabolic networks. It also acts as both curse and blessing engineers, where enzymes that have undergone directed may exhibit exquisite selectivity at expense diminished overall catalytic repertoire. We previously performed two independent campaigns on artificial leverages unique properties non-canonical amino acid (ncAA) para- aminophenylalanine (pAF) residue, resulting evolved variants which are specialized. Here, we combine mutagenesis, crystallography computation reveal molecular basis specialization phenomenon. In one variant, unexpected change quaternary structure biases substrate dynamics promote enantioselective catalysis, whilst other demonstrates synergistic cooperation between side chains pAF residue form semi-synthetic machinery. Our analysis provides valuable insights future engineering effective employ either widely used LmrR scaffold or residue.

Language: Английский

---