Machine Learning on Images of a Microbial Mutant Library

Author

Georgia Doing, Bard CollegeFollow

Date of Submission

Spring 2015

Academic Programs and Concentrations

Computer Science

Project Advisor 1

Brooke Jude

Project Advisor 2

Rebecca Thomas

Abstract/Artist's Statement

Environmental isolates, like BJB312, are interesting because of their potential therapeu- tic properties. Transposon mutagenesis is a technique used to determine the function of genes by randomly disrupting a genome and observing the phenotypic effects. The genome of BJB312 consists of over 5,000 genes, requiring 57,000 independent insertion mutants in order to break every gene in the genome. It is unwieldy to screen such a large library for defects. I used image processing techniques to convert qualitative data of mutant bacterial colonies morphology into a quantitative data set that is susceptible to data mining. Fur- ther, I built a tool of ensemble machine learning techniques that automatically analyze a large library of mutants. It first uses the unsupervised methods k-means and Wards hierarchical clustering to find a patterned, recurrent phenotype. It then uses a Support Vector Machine to screen the library at large. This tool is robust and useful on real-world data because it utilizes Machine Learning techniques to filter the image library before reaching the final clustering solution. Ten transposon insertion mutants that clustered together were characterized by lessened biofilm. This proof-of-concept study shows that genomic and high-throughput functional characterizations can be combined in order to rapidly explore a novel microbe.

Open Access Agreement

On-Campus only

Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Recommended Citation

Doing, Georgia, "Machine Learning on Images of a Microbial Mutant Library" (2015). Senior Projects Spring 2015. 102.
https://digitalcommons.bard.edu/senproj_s2015/102