Overview

We got in touch through WeChat with OUC-China in May, and we identified partnerships after a deep discussion with each other about one week later. Both our projects are related to the use of the selection markers to screen the strains during the upstream or downstream technologies of the synthetic biology. We have conducted in-depth cooperation in three aspects: experiment, modelling and education. (Click here to visit their wiki)

Experiment

During the experiments, our performance testing of the toxic genes was one step faster than OUC-China. In order to test the sacB gene in the eukaryotes and facilitate the process of OUC-China’s experiments in the meantime, we sent the plasmid harboring sacB gene and the relative information of the experimental conditions to OUC-China for further research.

As shown in figure 1A, OUC-China successfully introduced the 183 bp specific fragment of the sacB gene into Aureobasidium melanogenum.

As shown in figure 1B, the area ① was the one with the empty transformation plasmid, the area ② was the one with the sacB gene fragment transformation plasmid, and the area ③ was the one with 5 mM sucrose culture that was introduced with the sacB gene fragment transformation plasmid. We could see that the sacB gene has the toxicity of ontological expression and it did play a lethal effect when we added 5 mM sucrose. OUC-China verified that our inducible sacB also has toxic effects in eukaryotes to some extent, this preliminarily expanded the applicable objects of sacB.

Fig. 1. The results of the sacB gene used for OUC-China team. A：The gel picture for the plasmid with and without sacB gene fragment. B: The test for the toxic of sacB gene in Aureobasidium melanogenum.

Modeling

We also have common research interests in model construction.

In the early phase of modeling design, coding the DNA sequence, OUC-China used sequential coding and evaluated the frequency of gene sequence according to k-mer sequence. However, after several tests and comparison of the results, OUC found that the k-mers could not make the algorithm extract data features better and did not contribute much to improving the accuracy.

Both of our dry lab members communicated on this, trying to improve the accuracy of DNA sequence coding at the beginning of model construction. We suggested that we could add physicochemical descriptors to the training set. After intensive discussion, we both thought that the physicochemical descriptors can improve the performance of the model with great probability. Searching by our team members, we found a python package of DNA physicochemical descriptors, and shared with OUC-China to further study and test. After verification and comparison, their model improved the accuracy at the end. As we learn from figure 2, the green line (tolerance limit) is much closer to the best prediction when the model adding physicochemical descriptor (Figure 2A), which showed more accurate than previous version without physicochemical descriptor (Figure 2B).

There are several rounds of discussion related to physicochemical descriptors between OUC-China and us, which is very important and useful for our model construction.

Fig. 2. Random Forest Test Chart

When optimizing our respective models, OUC-China has relatively less cases of dataset. We suggested that they could abandon the strategy of the neural network and use some algorithms suitable for small data set training instead, such as random forest algorithm. OUC-China also suggested we can use grid search to determine the optimal super parameters of learners, which expanded our thoughts on modeling optimization.

Education

We invited OUC-China to participate in the discussion on education cooperation with NEU_CHINA, we discussed and shared ideas on education to expand our education forms and promote our understanding of publicity and education.

Our discussion mainly focused on the following three parts:

1. Understanding of synthetic biology.

2. Links between projects of each team and synthetic biology.

3. How to popularize synthetic biology.