Small Games
Inspired by the outstanding teams in the past, we have designed three small games in order to achieve the purpose of popularizing synthetic biology knowledge and let more people know and join us. Small game production is divided into two parts at the same time.
Inspired by the outstanding teams of previous years, we designed three mini-games to popularize the knowledge of synthetic biology and let more people know and join us. We first researched the relevant processes and regulations for mini-game production online, and only started production after we determined that it was safe, feasible, and non-infringing. The original goal was to make a biological version of the 2048 mini-game. The source code was found on the open source website GitHub as the original framework for development. After many discussions in group meetings, we designed the elements needed for the mini-game, paired with our team’s logo and cards to reflect our team’s style. After continuous modification and improvement, the biological version of the mini-game 2048 was completed. It was first trialled within the team, and then promoted through public accounts and tweets. It was then modified and improved based on the feedback received.
In the design stage, we used the WeChat mini game development platform to select the game template of Leap Rabbit and adapted it. However, during the production process, we found that due to the limitations of the original framework of the game, many elements of our team could not be integrated into the game. At the same time, the game loop lacked challenge during the clearance process. After consideration, we decided to develop a creature puzzle game while retaining the Jumping Rabbit mini-game. The latter has better integrated the elements of our team, and at the same time has a certain degree of challenge and fun, and can be used for PK with friends. After the game is completed, it will be tested within our team first. If there are no problems, it will be released to the open source community. After the production of the mini-game was completed, we wrote a copywriting summary, published and promoted it in the form of a public account, so that more people could participate and realize our original intention of designing the mini-game.
Hardware
Multifunctional microbial automatic culture machine
Design phase:
From a service point of view, in order to allow team members to better cultivate and detect microorganisms, we want to develop a related product. Through the learning and thinking of the team's own actual situation of the hardware BEST project in previous years, combined with the difficulties and needs encountered in the wet experiment, the team proposed a number of hardware project directions, and decided to develop a multifunctional microbial automatic culture machine in the process of discussion at several meetings, relevant literature learning, and exploration after practical testing. After determining the direction of the hardware project, we began to strengthen team building and design preparation in terms of learning and materials, and further discussion, refine the product development content, and develop an engineering practice plan, and design and manufacture this hardware product together with the recommendation of the team leader and the team.
Build phase:
According to the design planning flow chart, the team carried out detailed development division and stage acceptance in terms of theoretical construction and actual production. The product is divided into three parts: first, the realization of the functions of each module, including the selection and purchase of modules, code run-through, physical assembly, etc. Secondly, the design of the Internet of Things is carried out, including theoretical learning and comparative selection of Bluetooth or WiFi based communication and code function implementation. Finally, the shell design and module integration are carried out, including shell structure design, material selection, manufacturing process, module layout and assembly.
Test phase:
The test stage includes the function realization test of each module and the integration test of the finished product in two main stages. The stages are interconnected, so the error needs to often adjust the overall situation. The difficulty is not small, but in order to ensure product quality, serve the demanders, we adhere to repeated rigorous stage testing, step by step, the pursuit of better.
Learning Phase:
The engineering of this hardware product is a very practical learning process, involving a wide range of knowledge and high specialization. We need to continue to learn independently, give play to innovative ideas, strengthen information retrieval ability, hands-on production ability, team coordination and communication ability, etc., in short, in this process, we have gained a lot, it is a very meaningful learning stage.
Expected Phase:
At present, our products can basically meet the needs of the team's internal experimental personnel, but can not achieve our ideal effect, not enough for the broader, more professional group needs, so we expect to further upgrade in the future material selection, product function, quality and other aspects, so that the multi-functional microbial automatic culture machine can better play its utility for users.
Housing
Design phase:
In order to achieve the integration of the automatic training machine, we combined the target population, product functions, market demand and other factors to consider, decided to make a shell for the training mechanism.
Build phase:
By measuring the size of the specific parts, we first determined the actual inner shell size of the incubator, and also designed some grooves on the left, right and back three sides of the shell according to the actual situation to fix the inner plate. Then we use the drawing software to draw the specific model according to the preliminary hand-drawn sketch, in which we standardize the size and shape of the specific small parts, which can be used for 3D printing or to provide sheet cutting drawings for the factory. Finally, in order to enable customers to have a more intuitive understanding of our team's products, the model we drew was rendered.
Test phase:
We first contacted the project who could make the shell for us, provided them with detailed part drawings, and asked the factory to process according to the parts for us, and got the shell part. We then assemble the housing and parts, making small adjustments to the position of the internal support plate according to the specific situation.
Learning Phase:
According to the test, it is found that the temperature control effect of our products can not meet the expectations, and the problem may appear in the selection of shell materials. After that, we conducted market research again and consulted relevant literature to select some materials with better insulation effect. At the same time, we also considered the method of filling insulation materials between the inner shell.
MCF Skin worry-free web
design phase:
Starting from the perspective of better serving patients with skin diseases, after discussing and solving the actual situation, the team decided to develop a popular science dermatology platform. After having the general direction of development, the team further discussed, refined the platform development content, and formulated Developed a development arrangement plan, and connected with the team’s art team to design the page preview renderings of the platform.
Build phase:
According to the page preview renderings, the team has carried out detailed development division of labor, and the platform construction and the collection of popular science materials have been promoted simultaneously. The platform is built in two parts: static page display and specific functional modules. After setting up the platform framework, specific functions are embedded into the platform. Finally, popular science materials are updated simultaneously on the platform. The purpose of organizing popular science materials is to provide platform users with Popular science introduction to enrich the content of the platform. Through meetings and discussions within the group, the performance and details of the platform were further optimized, and the development of the platform was completed.
Testing phase:
The team held many discussions to test the operating performance and usage experience of the Skin Care Intelligent Platform, and further optimized the platform to better serve users.
Learning phase:
A meaningful popular science platform should not only disseminate popular science content, but also ensure the accuracy and timeliness of the popular science content, as well as user satisfaction of the platform. We update the latest science popularization consultation in real time and take user satisfaction evaluations seriously, and are committed to Further improve the popularization, timeliness and user-friendliness of the platform.
Prospective stage:
While the platform plays a role in popularizing science, it can develop other meaningful functions to make the platform more diversified and further improve user stickiness. We will maintain active and innovative research and development. Please continue to pay attention to our future research!
Hardware interactive software
In the later stages of designing and implementing a hardware product - a multi-functional automatic microbial culture machine, in order to give full play to the functions of the product itself and provide users with a more efficient and convenient culture experience, after team discussions and combined with the actual situation, we put the idea Transformed into an entity, I decided to design an IoT software that can be automatically controlled and intelligently interacted. I started several discussions with another member and initially determined the production direction of the software. Afterwards, the team members discussed with the team leader to further refine the software requirements and make the production direction clearer. After clarifying the production direction, I began to look for some available resources online and learn about software design and production. After communicating with the team members based on the available resources, it was finally determined that the software was a Bluetooth-based IoT method and named Zhipei.
After the preparation work was completed, I started software production. During the production process, I used the Bluetooth debugger to set module components suitable for the team's own hardware products, searched for and learned suitable control codes, and made many modifications, additions and deletions, constantly improving the code, and at the same time conducting actual control tests. When discovering problems Solve problems, conduct team discussions when encountering major difficulties, consult mentors, search for information, etc., and constantly make the software mature. During the process of software production and testing, I received many suggestions, such as: how to reasonably layout component locations, supplementary function settings, simple output feedback channel settings for code, etc. These internal testing suggestions have improved the software. After the internal testing function is completed, we will install this software on users' mobile devices and hope to get your experience and suggestions on this software. Fortunately, the software experience has been well received.
Result
Our original intention in making the software is to allow users to reduce unnecessary time in the process of cultivating bacterial strains and to better cultivate bacterial strains. Therefore, we have produced smart training software, which is a Bluetooth communication-type operating software. The key function settings in the software are in line with the module functions of a multi-functional automatic microbial culture machine. It can perform short-range control and visual reception of information, allowing cultivators to Enjoy the convenience of strain cultivation and allow users to obtain high-quality strain culture results.
The software interface has five functional modules, namely temperature control, timing, PH, lighting, and weight detection modules. After clicking the power button to turn it on, the user needs to select the module according to the required function. After selecting, they can perform specific operations such as start or pause, strengthen or weaken, etc. in the operation interface below. In addition, additional output control or information reception can be performed in conversation mode. The current shortcomings of this software are also relatively obvious. It cannot be used at long distances, the interface is not so beautiful, and some functions need to be matured. Therefore, in the future, we will prepare to upgrade to a software based on wife communication, and optimize the operation interface and improve it. Related functions.
Modeling Engineering
Our modeling was inspired by a paper published in Nature Communication, which reasonably used machine learning methods involving a machine learning framework, and applied it to the entire small molecule peptide sequence space to mine Potential antimicrobial peptides, we were extremely surprised to see this, because our experiments were very eager to find a method that can quickly and accurately find antimicrobial peptides. We immediately adjusted our research direction and started to learn machine learning-related knowledge from scratch, but Due to insufficient computer performance, we quickly gave up the work of reproducing the original paper, and instead adjusted the research purpose from mining the entire small molecule peptide sequence space to mining the candidate peptide sequence sample space. This not only improved the implementability of the solution sex, and more in line with the needs of biological workers. We learned the core idea in the original paper, which is to build a funnel-shaped machine learning pipeline. In terms of the specific design of the pipeline, we conducted more detailed explorations, discovered some shortcomings of the original solution, and made appropriate adjustments. Improvement, the specific plan is as follows:
1.1 Feature extraction
We found that the original solution incorporated a large number of feature encoding schemes, making the training features quite large, with a total of 676 features. We thought there might be some irrelevant redundant features, so we rebuilt the features. Specifically, We first tried the impact of different feature encoding schemes on model performance. We selected seven feature encoding schemes and used XGBoost as the classifier. The experimental results are shown in Figure 1. It can be found that AAC, CTDD, DP, DPC These four coding schemes have obvious advantages (the corresponding colors in the picture are darker), so we chose to fuse the above four coding schemes, but the number of features after fusion reached 635, which is not much different from the original scheme. Improvement, after analyzing the reasons, we found that there is still some redundant information between these four coding schemes, so we used F-score to calculate the ability of features to distinguish between the two categories, and finally selected 352 valid ones with higher quality. features, shortening the calculation amount by nearly 50%, and avoiding the impact of some redundant features on training, the performance of the model has been improved.
1.2 Classifier selection
Since the classification module handles the first step of the machine learning pipeline, the accuracy of its predictions is crucial. Misclassification errors in the first step may miss some very promising antimicrobial peptides, which is important to us. is unsatisfactory, so we chose to stack multiple base learners, and experimental verification showed that the stacked model has a certain improvement over the original model. Specifically, we tried a total of six machine learning methods: XGBoost, Random Forest (RF), Gradiant Boosting Decision True (GBDT), Logistic Regression (LR), Adaboost, and Decision True (DT). We found that The three methods have obvious advantages over the others, so we use them as the base learner, use stacking fusion technology, and use Logistic Regression as the second layer classifier, and finally get a model with better performance.
1.3 Introduce deep learning methods to improve the accuracy of regression modules
With the rapid development of artificial intelligence, more and more cases show that deep learning can better learn complex relationships and has greater development prospects, so we also tried to build different deep model frameworks ourselves. We initially built We have built a more complex deep neural network, but we found that no matter how we adjust the hyperparameters, the model always suffers from overfitting. Figure 3 shows some failure cases, from which we can see that with the training rounds With the increase, the model error always develops in unexpected directions.
After analyzing the reasons, we believe that our training data set is not difficult to learn, and only needs some relatively lightweight deep learning models to complete the task well, so we rebuilt the deep learning model. This time, the model Developing in a good direction, the prediction error shows an overall downward trend and gradually stabilizes after about 70 rounds, indicating that the model has completed learning.
Wet Engineering
Due to some force majeure factors, our experimental team experienced some personnel changes, and in the end only a few students were left persisting in conducting experiments. This undoubtedly had a certain impact on the progress of our experiments, making our progress slower than expected. At the same time, due to policy changes at Hainan University, our laboratory needed to move to a new campus. During this process, we encountered many unexpected difficulties, which made the conduct of experiments extremely difficult. Despite facing such huge challenges, our team did not give up. On the contrary, we actively seek solutions to overcome the impact of laboratory changes as much as possible so that subsequent experiments can proceed more smoothly. In order to better promote the experiment, we chose to enrich our experimental methods by reviewing relevant papers, asking teachers for advice, and using biological information, in order to achieve better results in the experiment.
For details, please click Moisture Experiment Project
