Halper, S. M. | Search bioRxiv

bioRxiv2020

The Synthesis Success Calculator: Predicting the Rapid Synthesis of DNA Fragments with Machine Learning

The synthesis and assembly of long DNA fragments has greatly accelerated synthetic biology and biotechnology research. However, long turnaround times or synthesis failures create unpredictable bottlenecks in the design-build-test-learn cycle. We developed a machine learning model, called the Synthesis Success Calculator, to predict whether a long DNA fragment can be readily synthesized with a short turnaround time. The model also identifies the sequence determinants associated with the synthesis outcome. We trained a random forest classifier using biophysical features and a compiled dataset of 1076 DNA fragment sequences to achieve high predictive performance (F1 score of 0.928 on 251 unseen sequences). Feature importance analysis revealed that repetitive DNA sequences were the most important contributor to synthesis failures. We then applied the Synthesis Success Calculator across large sequence datasets and found that 84.9% of the Escherichia coli MG1655 genome, but only 34.4% of sampled plasmids in NCBI, could be readily synthesized. Overall, the Synthesis Success Calculator can be applied on its own to prevent synthesis failures or embedded within optimization algorithms to design large genetic systems that can be rapidly synthesized and assembled.Competing Interest StatementThe authors have declared no competing interest.View Full Text

synthetic biology↗

bioRxiv2017

Automated Parameterization of Predictive Kinetic Metabolic Models from Sparse Datasets for Efficient Optimization of Many-Enzyme Heterologous Pathways

Engineered heterologous metabolic pathways can convert low-cost feedstock into high-value products, though it remains a significant challenge to reliably and efficiently maximize end-product biosynthesis, particularly when many enzymes must be co-expressed together. When current approaches are applied to many-enzyme pathways, the construction and characterization process is highly iterative and laborious, while generating high-dimensional datasets that remain difficult to analyze for forward engineering efforts. To overcome these challenges, we developed a new algorithm that determines the highly non-linear and high-dimensional relationship between a pathways enzyme expression levels and its end-product productivity from common characterization of a small number of heterologous pathway variants. We combined kinetic metabolic modeling, elementary mode analysis, model reduction, de-dimensionalization, and genetic algorithm optimization into an automated procedure that parameterizes accurate kinetic metabolic models from sparsely characterized pathway variant libraries with varied enzyme expression levels. The resulting Pathway Maps are used to determine rate-limiting steps, predict optimal expression levels, identify allosteric interactions, rank-order enzyme kinetics, and prioritize protein engineering efforts. We demonstrate the Pathway Map Calculator algorithm on two experimental datasets, a 3-enzyme carotenoid biosynthesis pathway and a 9-enzyme limonene biosynthesis pathway, as well as a series of in silico pathway examples to rigorously demonstrate the algorithms accuracy, linear scaling, and high tolerance to measurement noise. By greatly reducing experimental efforts and providing quantitative forward engineering predictions, the Pathway Map Calculator has the potential to dramatically accelerate the engineering of many-enzyme heterologous metabolic pathways.\n\nHighlightsO_LIWe developed an automated algorithm that uses a small number of characterized pathway variants to determine the pathways expression-productivity relationship.\nC_LIO_LIThe Pathway Map Calculator is accurate, scales linearly on many-enzyme pathways, distinguishes allosteric interactions, and tolerates substantial measurement noise.\nC_LIO_LIPathway Maps are used to predict optimal enzyme expression levels, identify rate-limiting steps, and prioritize protein engineering efforts\nC_LI

synthetic biology↗