Skip to content

Proposal.md

Latest commit

 

History

History
38 lines (27 loc) · 4.45 KB

Proposal.md

File metadata and controls

38 lines (27 loc) · 4.45 KB
Raw

Bio

I am Gates Kempenaar and I am a 3rd year Computer Science student at the University of Lethbridge. During this term I will be developing tools to predict minimum inhibitory concentration (MIC) for thirteen different antimicrobial agents using genome sequences. I will focus on Salmonella enterica serovars Kentucky and Heidelberg, and test how dataset composition affects performance.

Introduction

S. enterica is a food-borne pathogen that causes gastrointestinal infections such as: typhoid fever, paratyphoid fever, and salmonellosis. In Canada, in 2016 alone, an estimated 87,500 people contracted Salmonella spp., resulting in 925 hospitalizations and 17 deaths (3).

The Heidelberg and Kentucky serovars are most commonly found in North America. In general, bacteria of these serovars have increased antimicrobial resistance (AMR) compared to bacteria of other serovars of S. enterica, and are often multiple-drug resistant (MDR) (1,2,4). The amount of antimicrobial agent required to inhibit bacterial growth is refereed to as the MIC. MIC values are currently determined using laboratory tests that are costly and time consuming (5). Whole genome sequencing (WGS) of bacteria is becoming routine for both surveillance and outbreak response, and we aim to directly identify MIC values from genome sequence. This approach has previously been used for the bacterial pathogen Klebsiella pneumoniae, where machine learning models could accurately predict MIC values within 1 dilution-factor >90% of the time (11).

In this coop term I will be using publicly available, and Canadian WGS data sets generated as part of the Genomics Research and Development Initiative (GRDI) on AMR, with their accompanying MIC values. I will be focusing on the S. enterica serovars Heidelberg and Kentucky because the provided data sets contain the largest amount of genome data.

The challenge that I will be exploring is how diversity among datasets affects prediction accuracy between datasets; the Canadian Kentucky and Heidelberg genomes are less diverse in genome content than the publicly available data.

By using two separate data sets of varying diversity, we hope to identify the genomic features responsible for differences in MIC, rather than genomic regions that are merely correlated with the differences. This will allow better models to be built that are capable of making better MIC predictions for these important serogroups in the future.

Implementation

WGS data will be reformatted for use in the supervised machine learning models that will be used to predict MIC values.

I will be using machine learning models because they are fast, scalable, and only require WGS data and corresponding MIC values to be able to make a prediction. The raw WGS data will be converted to a numeric input based on the frequency of DNA sequence substrings of size "k", called k-mers. Counts of k-mers with length 11 will be computed for each genome using the program Jellyfish (6). A k-mer length of 11 was previously found to be optimal.

We will use two machine learning methods: Gradient Boosted Decision Trees (XGBoost) (7) and artificial neural networks (ANN). XGBoost uses an ensemble of weaker decision trees that when working together to answer a series of true or false questions create a strong model to make a prediction. An ANN uses layers of neurons with corresponding weights to make a prediction based on specified input neurons. The XGBoost model will be used as implemented using the XGBoost Python package and the ANN using Keras (8), TensorFlow (9), and scikit-learn (10).

Conclusion

Successful completion of this project will lead to faster and more accurate MIC predictions based solely on genome sequence data. This could decrease the cost and reduce testing times of antimicrobial resistance in Salmonella enterica, leading to healthier Canadians.

References

  1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5047448/
  2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5622158/
  3. https://www.canada.ca/en/public-health/services/food-borne-illness-canada/yearly-food-borne-illness-estimates-canada.html
  4. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1082862/
  5. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4503790/
  6. https://github.com/gmarcais/Jellyfish
  7. https://xgboost.readthedocs.io/en/latest/
  8. https://keras.io/
  9. https://www.tensorflow.org/
  10. http://scikit-learn.org/stable/index.html
  11. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5765115/