An Automated Model Test System for Systematic Development and Improvement of Gene Expression Models

Abstract

Gene expression models greatly accelerate the engineering of synthetic metabolic pathways and genetic circuits by predicting sequence-function relationships and reducing trial-and-error experimentation. However, developing models with more accurate predictions is a significant challenge, even though they are essential to engineering complex genetic systems. Here we present a model test system that combines advanced statistics, machine learning, and a database of 9862 characterized genetic systems to automatically quantify model accuracies, accept or reject mechanistic hypotheses, and identify areas for model improvement. We also introduce Model Capacity, a new information theoretic metric that enables correct model comparisons across datasets. We demonstrate the model test system by comparing six models of translation initiation rate, evaluating 100 mechanistic hypotheses, and uncovering new sequence determinants that control protein expression levels. We applied these results to develop a biophysical model of translation initiation rate with significant improvements in accuracy. Automated model test systems will dramatically accelerate the development of gene expression models, and thereby transition synthetic biology into a mature engineering discipline..

Automated Model-Predictive Design of Synthetic Promoters to Control Transcriptional Profiles in Bacteria

“…We designed the 14206 promoter sequences to measure how each motif sequence alters these free energies to create a sequence-complete model, including all possible -10 hexamers, all possible - 35 hexamers, all possible -10 extended motifs within 8 consensus and anti-consensus background configurations, 229 sequence spacers with varied lengths and nucleotide compositions, 2420 UP elements with varied AT content within 4 consensus and anti-consensus background configurations, 735 discriminator elements with varied lengths and GC content, and 582 ITR sequences with varied R-loop stabilities ( Figure 1B ). Using oligopool synthesis and two-step library cloning, we constructed a barcoded plasmid pool that uses each promoter in a common genetic context, expressing a single protein with a moderate translation initiation rate (about 5000 on the RBS Calculator v2.1 scale 5 ), with unique barcode sequences positioned in the 3’ untranslated region to avoid confounding effects.…”

“…Transcription is the gene expression process responsible for producing all RNA and is a common engineering target for creating novel products, including microbial chemical factories, toxinsensing genetic circuits, and mRNA vaccines [1][2][3] . However, while DNA assembly techniques enable the construction of custom-designed genetic systems 4 , it remains challenging to a priori predict and control a system's gene expression profile 5 , for example, by initiating transcription with desired rates at specific DNA start sites, while minimizing transcription from all other DNA sequence regions. Currently, transcriptional control relies on empirical characterization of promoters as modular genetic parts 6 .…”

“…In state 2, RNAP/s is bound to the promoter DNA with a stable transcriptional bubble, which includes the transition from the closed-to-open conformation, the initial melting of promoter DNA to create an unstable transcriptional bubble, and the formation of a stable transcriptional bubble (R-loop) via initial transcription of the ITR region. The transitions and transition rates in this system include: [1] the association rate of RNAP/s recruitment to promoter DNA, transitioning the system from state 0 to 1 (r1); [2] the disassociation rate of RNAP/s after it's bound, transitioning the system from state 1 to 0 (r-1); [3] the rate of the closed-to-open conformational change, DNA melting, and stable R-loop formation, transitioning the system from state 1 to 2 (r2); [4] the rate of transcriptional bubble collapse and abortive initiation, transitioning the system from state 2 to 1 (r-2); and [5] the rate of productive transcriptional initiation, transitioning the system from state 2 to 0, whereby RNAP/s transitions to the transcriptional elongation process, escapes from the promoter DNA, and frees up the promoter DNA to be available for binding to another copy of RNAP/s (r3).…”

Tuning Cell-free Composition Controls the Time-delay, Dynamics, and Productivity of TX-TL Expression

“…We therefore augmented the RBS Calculator model to explicitly include the apparent translation elongation rate of the protein’s coding sequence. As the starting point, the RBS Calculator calculates the ribosome’s binding free energy (ΔG total ) using a 5-term free energy model 20, 21, 25, 26 and then predicts a protein coding sequence’s translation initiation rate (r init ) according to: where the 30S ribosomal subunit’s binding free energy (ΔG total ) depends only on the mRNA’s sequence and c 1 is a proportionality constant that accounts for extrinsic differences here influenced by the cosolute composition. After the 30S ribosomal subunit binds to the mRNA, it recruits the 50S ribosomal subunit, forms the 70S initiation complex, and initiates translation.…”

Development and comparison of cell-free protein synthesis systems derived from typical bacterial chassis

“…Different RBS sequences were designed for each CFPS system to replace the original RBS sequence between the T7 promoter sequence and the initiation of sfGFP gene. RBS sequences were designed according to different hosts by the RBS library calculator (De Novo DNA: RBS Library Calculator) (Reis and Salis 2020 ; Salis et al 2009 ). Because the range of translation initiation rates were very wide, so six RBS sequences (Table 1 ) were equably selected from the minimum to the maximum translation initiation rates.…”