Active and machine learning-based approaches to rapidly enhance microbial chemical production

Kumar, Prashant; Adamczyk, Paul; Zhang, Xiaolin; Andrade, Ramon Bonela; Romero, Philip A.; Ramanathan, Parameswaran; Reed, Jennifer L.

doi:10.1016/j.ymben.2021.06.009

Cited by 16 publications

(12 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Progress in well-established high-throughput technologies and automation primarily benefited the acquisition of large datasets [26,27], which are essential for ML. The previous studies demonstrated that introducing ML to medium optimization successfully accelerated bacterial growth [28][29][30] and increased productivity [31,32].…”

Section: Introductionmentioning

confidence: 99%

Machine learning-assisted medium optimization revealed the discriminated strategies for improved production of the foreign and native metabolites

Aida

Uchida

Nagai

et al. 2023

Preprint

View full text Add to dashboard Cite

The medium composition was crucial for achieving the best performance of synthetic construction. What and how medium components determined the production of the synthetic construction remained poorly investigated. To address the questions, a comparative survey with two genetically engineered Escherichia coli strains was performed. As a case study, the strains carried the synthetic pathways for producing the aromatic compounds of 4APhe or Tyr, which were common in the upstream but differentiated in the downstream metabolism. Bacterial growth and compound production were examined in hundreds of medium combinations that comprised 48 pure chemicals. The resultant data sets linking the medium composition to bacterial growth and production were subjected to machine learning for improved production. Intriguingly, the primary medium components determining the production of 4PheA and Tyr were differentiated, which were the initial resource (glucose) of the synthetic pathway and the inducer (IPTG) of the synthetic construction, respectively. Fine-tuning of the primary component significantly increased the yields of 4APhe and Tyr, indicating that a single component could be crucial for the performance of synthetic construction. Transcriptome analysis observed the local and global changes in gene expression for improved production of 4APhe and Tyr, respectively, revealing divergent metabolic strategies for producing the foreign and native metabolites. The study demonstrated that ML-assisted medium optimization could provide a novel point of view on how to make the synthetic construction meet the original design.

show abstract

Section: Introductionmentioning

confidence: 99%

Machine learning-assisted medium optimization revealed the discriminated strategies for improved production of the foreign and native metabolites

Aida

Uchida

Nagai

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Beyond that, metabolic engineering equipped with machine learning techniques (neural network and Bayesian optimization etc.) ( 99 , 100 ), and omics analysis ( 101 , 102 ) could also be employed to design, regulate, and optimize the metabolic pathway for high-efficiency LCPUFAs rich oil production. In the future, comprehensive interdisciplinary research will become the theme and contribute to enzymatic research.…”

Section: Discussionmentioning

confidence: 99%

Key Enzymes in Fatty Acid Synthesis Pathway for Bioactive Lipids Biosynthesis

et al. 2022

View full text Add to dashboard Cite

Dietary bioactive lipids, one of the three primary nutrients, is not only essential for growth and provides nutrients and energy for life's activities but can also help to guard against disease, such as Alzheimer's and cardiovascular diseases, which further strengthen the immune system and maintain many body functions. Many microorganisms, such as yeast, algae, and marine fungi, have been widely developed for dietary bioactive lipids production. These biosynthetic processes were not limited by the climate and ground, which are also responsible for superiority of shorter periods and high conversion rate. However, the production process was also exposed to the challenges of low stability, concentration, and productivity, which was derived from the limited knowledge about the critical enzyme in the metabolic pathway. Fortunately, the development of enzymatic research methods provides powerful tools to understand the catalytic process, including site-specific mutagenesis, protein dynamic simulation, and metabolic engineering technology. Thus, we review the characteristics of critical desaturase and elongase involved in the fatty acids' synthesis metabolic pathway, which aims to not only provide extensive data for enzyme rational design and modification but also provides a more profound and comprehensive understanding of the dietary bioactive lipids' synthetic process.

show abstract

“…In these settings, predictive modeling has traditionally been performed by relatively data-free biophysics-based modeling, such as the Rosetta energy function . Increasingly, as laboratory techniques for property measurement improve in cost and scale, data-hungry machine learning (ML)-based predictive modeling has started to supplant biophysics-based approaches, owing to improved performance. − Indeed, on test molecules that lie “near” the training data, ML models are often more accurate than biophysics-based models that rely on approximations of the underlying physics for computational tractability. However, the number of labeled data is typically minuscule compared with the size of the protein/molecule space for which one seeks to make predictions.…”

Section: Introductionmentioning

confidence: 99%

Coherent Blending of Biophysics-Based Knowledge with Bayesian Neural Networks for Robust Protein Property Prediction

Nisonoff,

Wang,

Listgarten

2023

ACS Synth. Biol.

View full text Add to dashboard Cite

Predicting properties of proteins is of interest for basic biological understanding and protein engineering alike. Increasingly, machine learning (ML) approaches are being used for this task. However, the accuracy of such ML models typically degrades as test proteins stray further from the training data distribution. On the other hand, models that are more data-free, such as biophysics-based models, are typically uniformly accurate over all of the protein space, even if inferior for test points close to the training distribution. Consequently, being able to cohesively blend these two types of information within one model, as appropriate in different parts of the protein space, will improve overall importance. Herein, we tackle just this problem to yield a simple, practical, and scalable approach that can be easily implemented. In particular, we use a Bayesian formulation to integrate biophysical knowledge into neural networks. However, in doing so, a technical challenge arises: Bayesian neural networks (BNNs) enable the user to specify prior information only on the neural network weight parameters, rather than on the function values given to us from a typical biophysics-based model. Consequently, we devise a principled probabilistic method to overcome this challenge. Our approach yields intuitively pleasing results: predictions rely more heavily on the biophysical prior information when the BNN epistemic uncertainty�uncertainty arising from a lack of training data rather than sensor noise�is large and more heavily on the neural network when the epistemic uncertainty is small. We demonstrate this approach on an illustrative synthetic example, on two examples of protein property prediction (fluorescence and binding), and for generality on one small molecule property prediction problem.

show abstract

Active and machine learning-based approaches to rapidly enhance microbial chemical production

Cited by 16 publications

References 41 publications

Machine learning-assisted medium optimization revealed the discriminated strategies for improved production of the foreign and native metabolites

Machine learning-assisted medium optimization revealed the discriminated strategies for improved production of the foreign and native metabolites

Key Enzymes in Fatty Acid Synthesis Pathway for Bioactive Lipids Biosynthesis

Coherent Blending of Biophysics-Based Knowledge with Bayesian Neural Networks for Robust Protein Property Prediction

Contact Info

Product

Resources

About