Frontpage Data visualisation Parametizing data Directory structure R-package SQL Zotero Reproductibility Future endeavours Free research (Machine learning) CV Bibliography
For my free room, I was given 4 days (32 hours) to research a data-science subject of choice. As my interests lie within microbiology, I went to search for a (for me) new skill in data sciences which seemed to have much use in microbiology. I chose machine learning, as it seemed to be relevant in most if not all microbiology-related research, from antiobitic resistance to microbiota research to automated agar plate-based bacteria identification.
I originally going specifically into image-based machine learning for atomating bacterial identifcation, however, I was immediately confronted by how horrendously difficult that would be. Thus, I decided to instead switch to some more basic techniques, like KNN-based machine learning. Then, I looked for an more realistic end-goal than an image-based algorythm, and thought that being able to use a algroythm like IDTAXA would be a good, concrete end goal.
I think that during these 4 days of time, I’ve definitely learned the basics of machine learning. Getting a feel for two flexible algorythms and 1 pre-made algorythm, learning the basics of how to set up a algorhythm, learning some forms of pre-processing, I’ve definitely learned a about machine learning in this time. Although I most definitely still could not yet apply this knowledge on a research project, I do now have a basis which I could use to, under guidance of a more advanced data scientist, set up a practical machine learning algorythm.
If I were to re-do this entire machine-learning research project, there’d be 2 things I’d definitely want to change.
1: Pick a different mock dataset than Glass. Glass is small, has unbalanced test conditions, and it’s datapoints aren’t really intuitive. If would’ve known beforehand that I was gonna use it mutliple times, as an example for all of my skills, I would’ve definitely chose a microbiology-related dataset instead.
2: Dedicate more time to using machine learning on genuine studies. I was originally planning on performing the IDTAXA determination on a study using 16s-RNA to study the microbiome of professional athletes, however, due to SRA-toolkit not functioning within R-studio and it for some reason not working easily on a loose command prompt too, this could not be done within the time limit of the research.
Despite these points of improvement, I’m definitely still satisfied with what I’ve learned.