Team:KUAS Korea/HumanPractices/IntegratedPractices

Integrated Practices

Consulting with the computer modeling and Datamining expert

While seeking methods to analyze our team’s wet lab data and building significant models, we insisted that there should be further validations to make a more fitting model and its formula. Therefore, we consulted on modeling methods and how to validate our modeling process. Since our project is based on the analyzing of the lab experiment data, we approached this through a top-down method and tried to figure out the formula for the modeling. The problematic situation we faced was having only little data to use as a source of modeling. On this problem, he advised us to approach in a more intuitive manner when modeling in a situation with a small dataset. The validation method he recommended was the hold-out method on history data and use the N fold cross-validation. Since our data is especially in small number, he advised us to use the Leave-one-out cross-validation for better modeling. Furthermore, he claimed that making a sophisticated modeling that perfectly explains the experiment results may not be the most general and all-round modeling. He asserted that we should focus on simplifying the modeling formula via applying the Ockham’s razor so that we could grasp the key signals and exclude the noises.

In addition, upon our opinion of expanding behavior types of the microorganisms, he told us that the modeling would become more sophisticated and we might have to divide cases to visualize into a significant graph. Finally, we asked him about the effectiveness of our research on other fields of studies. Since our experiment condition is very restricted, he explained that our project would only be applicable in certain situations and we should expand the uses by specifying it to certain situations that might require this kind of technique.

Thanks to the interview, we could redress the situation of a plateau in the modeling process and find new ways to validate our experiment results which are apparently small in number. Moreover, we insisted that to enlarge the use of our project, we should add certain behaviors that are actually detected in nature and explain the survival rate ratio between different types of microorganisms in coculture environments.