Team:SYSU-CHINA/Collaborations

2. 2020 iGEM Southern China Regional Online Meeting

For us, iGEM is not only a competition to gain honors, but also a process of meeting like-minded friends, and a process of win-win cooperation. On this way forward, we are happy that there are two outstanding teams walking side by side with us and making progress together. What's more, we attended this year's CCiC（Conference of China iGEMer Community）and iGEM Southern China Regional Online Meeting, from which we got plenty of valuable advice.

Figure 1. Collaborate with SJTU-BioX Shanghai

1. Cooperated With SJTU-BioX Shanghai

We are very fortunate to get acquainted with SJTU-BioX-Shanghai and establish a long-term and in-depth cooperative relationship with them. In the process of cooperation, both of us have kept communicating, making progress, and moving towards a better goal!

Although due to the epidemic, we and SJTU-BioX-Shanghai could not meet offline, but this didn't stop our enthusiasm of cooperating with each other. We have cooperated with SJTU-BioX-Shanghai in various aspects through online conferences. We have conducted many online exchanges in many aspects including modeling, algorithm and education. Both of us analyze each other's projects, ideas, methods and shortcomings, and do our best to help each other to improve our projects.

Model Parts

In the modeling part,We not only discussed the model construction of the two teams in detail, but also provided sincere suggestions and important help for each other's project model design details.

Figure 2. Our online meeting on modeling parts

In the modeling cooperation with SJTU-BioX-Shanghai, we told them about our ideas for experimental modeling. Based on the actual situation of our experiment, they gave us some targeted suggestions. With their suggestions, we have made major changes to the overall idea and specific implementation of modeling.
Our original idea was:

Directly use the data that we can get in the experiment to fit the relationship between the ability of dsRNA to inhibit ADAR1 and the survival rate under different IFN and Dox concentrations. However, the "inhibition capability of dsRNA to ADAR1" we want cannot be obtained directly from experiments, so we define a value "evolution percentage" as the ratio of the current dsRNA's inhibition capability to the ideal inhibition capability.

Figure 3. Parts of our modeling design

The "inhibition capability" is quite abstract. Therefore, we can fix a relatively high concentration of IFN and Dox according to our own needs. Under these conditions, the inhibition capability of dsRNA that can make the cell survival rate reach the specified value is ideal. At the same time, we consider the evolution percentage of dsRNA in this state is 1.

The functional relationship between the affinity of dsRNA and it’s inhibition capability is not clear to us, but if it is roughly considered to be close to linner, then the value of the evolution percentage is approximately equal to the ratio of the affinity to the ideal affinity. In this way, the data we use for fitting are all available from the experiments.

The next step is to carry out directed evolution. Through the data obtained in the experiment, we can gradually make our model more substantial and more perfect.

We chose this idea at that time, because although this design does not seem to be very complicated, nor does it calculate the mechanism and corresponding relationship of each step, but everything is based on the input of the experiments and the data obtained from the experiments, which is more suitable for the experiment. It can also gradually grow with the progress of the experiment. The defined quantity "evolution process" and this set of calculation methods can also be applied to other directed evolution experiments. Therefore, it has a high universality.

However, the students from SJTU-BioX-Shanghai pointed out that due to the special circumstances this year, the experiment time is relatively short, and our model relies heavily on experiments, so the effect obtained by this model is probably not very good. Under their suggestions, we chose a more classic pathway modeling, which can explain the principle of our pathway more rationally, and introduce the feasibility of our model more clearly in mathematics.

Algorithm Parts

In terms of algorithms, we have had many in-depth exchanges and benefited a lot from each other.

Figure 4. Deep discussion in algorithms

After several online meetings, we pointed out each other's possible shortcomings and made certain improvements.

First of all, we made suggestions for the data set division of the SJTU machine learning part. The magnitude of the positive and negative training sets of this module differs by 10^2, which would cause the corresponding loss of the negative sample to dominate, thus affecting the judgment of the machine learning model , so we suggested them to increase the number of positive sample training sets, but considering the limited experimental data, it was impossible to increase the positive sample training set, then we could consider the regularization method to adjust the training model to improve the fitting effect of the model. Or by adjusting the sample weight to improve the model, and then improve the model fitting effect. Other suggestions included adding additional analysis of positive samples that were not included in the prediction, and conducting some research on whether there were key bases in the sequence.

Similarly, we were also very grateful to SJTU for their valuable suggestions on algorithm.As for the module of our neural network, SJTU proposed that the regression model we used to analyze RNA sequence and affinity data would be difficult to train. This model could be changed to a classification model. In fact, we did the same; for our feature extraction, It was believed that the possible elements would be too single. If the number of features were increased, the accuracy of the prediction result might be improved; finally, after we used the classification method, there was a problem of over-fitting and the loss of the test set is not significantly reduced. They recommended that we improve Network structure, such as adding dropout layer after full-connected layers or regularization, or try other optimizers, such as Adam, to reduce the learning rate during training and adjust it according to the accuracy of the result. When conditions permit, we have made considerable efforts to try and improve each other.

Education Together

Our two team cooperated to carry out popular science education with the theme of directed evolution. The two sides jointly wrote plans and prepared popular science materials for the Guangdong Experimental High School and High School Affiliated to Fudan University students and their IGEM team. Due to the epidemic, this popular science exhibition was carried out in the form of online conferences, divided into Guangdong Experimental High School and High School Affiliated to Fudan University. During the popular science exhibition, the two parties cooperated to complete the display, preaching the knowledge of directed evolution and their respective iGEM projects this year. And then will have a vocal discussion with high school students to answer questions raised by students.

Figure 5. A speech on directed-evolution

2. 2020 iGEM Southern China Regional Online Meeting

In June, we participated in the 4th Southern China Regional Meeting held by SZU-China. In all, about 25 Chinese teams took part in the exchange. Due to the impact of the epidemic, this year's meeting could not be held offline, but only online presentation and communication were permitted. Although there was no face-to-face contact, online interaction had its unique advantages, such as allowing us to exchange ideas more freely and made it possible for us to connect with a wider range of teams.

During the meeting, we not only learned interesting and novel ideas from various teams, but also listened to the lectures from guests invited by the conference. One of the biggest things we’ve known was how to organize our work online using various kinds of software, which was significant because COVID-19 prevented us from returning to campus. Also, after the meeting，we received a beautiful trophy as souvenir from our friends of SZU-China. Thanks a lot!

Click here to see how SZU-China described the meeting.

Figure 8. Souvenir of 4th Southern China Regional Meeting

3. 7th CCiC Meetup

On August 28 to 31, we participated in the 7th CCiC (Conference of China iGEMer Community) online due to the affect of COVID-19. This year, its theme is from LAB (laboratory) to FAB (fabrication).

CCiC is a regional meeting co-sponsored by iGEMer in China, and is designed to provide a wide platform for the exchanges of different iGEM teams. It's like a preview of Giant Jamboree, which allows teams to show their projects and posters, learning from each other and giving comments and suggestions.

During CCiC, we’ve received a lot of helpful suggestions and reviews, including how to improve the project, way to make our presentation and the poster design. Specially, we obtained advice and guidance from Senior Jianzhao, Yang on quantitation of selection pressure in directed evolution, which is of great significance to our follow-up modeling work. We also received some lovely souvenirs provided by CCiC committee. Thanks a lot!

Figure 6. Screen shoot of 7th CCiC

Figure 7. Souvenirs from CCiC committee