Quantitative design of regulatory elements based on high-precision strength prediction using artificial neural network

Abstract

Accurate and controllable regulatory elements such as promoters and ribosome binding sites (RBSs) are indispensable tools to quantitatively regulate gene expression for rational pathway engineering. Therefore, de novo designing regulatory elements is brought back to the forefront of synthetic biology research. Here we developed a quantitative design method for regulatory elements based on strength prediction using artificial neural network (ANN). One hundred mutated Trc promoter & RBS sequences, which were finely characterized with a strength distribution from 0 to 3.559 (relative to the strength of the original sequence which was defined as 1), were used for model training and test. A precise strength prediction model, NET90_19_576, was finally constructed with high regression correlation coefficients of 0.98 for both model training and test. Sixteen artificial elements were in silico designed using this model. All of them were proved to have good consistency between the measured strength and our desired strength. The functional reliability of the designed elements was validated in two different genetic contexts. The designed parts were successfully utilized to improve the expression of BmK1 peptide toxin and fine-tune deoxy-xylulose phosphate pathway in Escherichia coli. Our results demonstrate that the methodology based on ANN model can de novo and quantitatively design regulatory elements with desired strengths, which are of great importance for synthetic biology applications.

Figure 1. Relative strengths of the constructed Trc promoter & RBS library.

The region of Trc promoter & RBS in pTrcHis2B is selected for random mutagenesis by error prone PCR, and mutants with various strength are obtained by detecting the fluorescent intensity of GFP after screening by 48-deep-well plates and flow cytometry assay.

Figure 2. Functional relationship between the prediction performance of ANN models and the scale of training data set.

Training data set scale ranges from 40 to 90 sequences. (A) Maximum and minimum SSE values of prediction as a function of training data set scale. (B) Maximum and minimum prediction errors as a function of training data set scale.

Figure 3. The well trained BP-ANN model NET90_19_576 can finely predict the measured Trc promoter & RBS strengths.

(A) The predicted relative strengths of promoter & RBS fit with the measured values using the data of training set. (B) The predicted values fit with the measured values using the data of test set. (C) The comparison results between prediction values and target values (experiment values). (D) The best fitting results of log Trc promoter & RBS activities with their PWM scores.

Figure 4. Effect of each single point mutation on sequence strength and sequence conservative analysis.

(A) Sequence strength influenced by mutation of each single site. Red indicates positive mutation while blue indicates the negative. Deeper color means more significant change of strength. Each box represents one base in the sequence. Figure in the boxes is the location number of this base, while the subscript indicates that this base is mutated to another one (e.g., A→C means A mutated to C, and T→G means T mutated to G, etc.). (B) Conservative analysis of high activity sequences (strength >1). Bases in the boxes are conservative points. ‘+/−’ indicates this point is predicted to be a positive/negative ‘key-point’. Same as below. (C) Conservative analysis of extremely low activity sequences (strength <0.1). The analysis was performed using online WebLogo Tool (http://weblogo.threeplusone.com/create.cgi). (D) Count of mutation types of the ‘key-points’. Figure in the boxes is the count of negative or positive mutation number.

Figure 5. Promoter & RBS sequence design based on ANN prediction model.

(A) Sequence with desired strength can be designed by the following strategies: i) 8 out of 10,000 sequences (s01–s08) are randomly selected from an in silico Trc promoter & RBS library generated based on ANN predicting model NET90_19_576; ii) sequences (s11–s15) with desired strength can be generated by repeated introduction of random mutations into the wild-type sequence under a certain mutation rate; iii) sequences (s21–s23) with desired strength can be generated by using different combinations of ‘key site’ mutations based on the prediction of NET90_19_576. All designed sequences were synthesized and their strengths were tested and compared with the design strength.

Figure 6. Application of designed elements for peptide BmK1 expression and DXP pathway engineering in E. coli.

(A) Sketch maps of plasmids for designed elements applications. Plasmids s21-gfp, s05-gfp and s14-gfp contain gene gfp between BamHI/EcoRI sites, plasmids s21-bmk1, s05-bmk1 and s14-bmk1 contain gene bmk1 between NcoI/HindIII sites, plasmids s21-dxs, s05-dxs and s14-dxs contain gene dxs between NcoI/EcoRI sites. (B) Effect of applying designed elements for peptide BmK1 expression and DXP pathway engineering in E. coli. The wild-type Trc promoter and RBS (without inserting dxs gene) served as the blank control.