Using Machine Learning to Predict AMR

Love Tsai, Computer Science, July 19, 2020

A researcher has inoculated two agar plates with particular microbes, both treated with drugs to test their antimicrobial resistance (This procedure is called the Kirby-Bauer test). The left disc shows a normal response—clear rings indicating that bacteria has not grown—but the right disc shows almost unaltered growth despite exposure to bacteria-deterrent drugs. This phenomenon, termed anti-microbial resistance, is a looming problem facing scientists and public health in the 21st century. 

Source: Wikimedia Commons

Humanity’s first forays into the world of antibiotics began with the discovery of the syphilis treatment arsphenamine in 1909 by Paul Ehrlich.1 Not long after, Alexander Fleming discovered penicillin by chance in 1928. The discovery of penicillin was especially groundbreaking because it could be used to treat many different types of infections such as those caused by Listeria and streptococci.1 It was later used extensively during World War II to prevent open wounds from developing into fatal infections.3 Fleming was awarded a Nobel Prize for his efforts, during which he gave the crowd a warning that now rings all too familiar:

“It is not difficult to make microbes resistant to penicillin in the laboratory by exposing them to concentrations not sufficient to kill them, and the same thing has occasionally happened in the body. The time may come when penicillin can be bought by anyone in the shops. Then there is the danger that the ignorant man may easily underdose himself and by exposing his microbes to non-lethal quantities of the drug make them resistant.” 4

In the quote above, Fleming spoke of the possibility that penicillin would be rendered less effective by penicillin-resistant bacteria (which can be developed when not enough of the drug is used during treatment to kill the bacteria but is still absorbed and recognized). Today, drug resistance for all antibiotics, not just penicillin, is a major concern across the pharmaceutical industry. This is because antimicrobial resistance (AMR) can lead to untreatable superbugs that render medical intervention ineffective. AMR is a natural process stemming from principles of evolution: as survival pressures affect a population of bacteria, those that endure the longest possess some sort of biological advantage against the stressor. Because these evolutionary fit, antibiotic-resistant bacteria survive and reproduce, AMR develops over several generations.6

All bacteria can become antibiotic resistant, some even to common, alcohol-based hand sanitizers. The ability to track, predict, and monitor such microbes is becoming increasingly important every day, but it can be difficult to sort through, identify, and analyze the millions of organisms that coexist with humans. Traditionally, researchers have screened the genes of bacterial species to detect sequences indicative of AMR. However, this method is tedious and unreliable in the face of newly sequenced genomes where there is no established pattern to look for. As such, researchers at Washington State University have developed a machine-learning software program aimed to detect these AMR bacteria without the need for a reference genome.2

The WSU team used a machine-learning (ML) algorithm informed by game theory that analyzes features of AMR proteins.2 Machine learning is designed to extract useful information from complex data sets. A common algorithm is linear regression. Other types include decision trees, k-means analysis, and random forests.5 Game theory goes hand-in-hand with ML when researchers are interested in strategic actions between players (in this case, the strategic, evolved features that can lead to AMR). This approach allows the researchers to look at various interactions, similarities, and differences between these organisms’ individual characteristics, not just their genotypes.2 The program, which is packaged in a Graphical User Interface (GUI) frame for easy use, gives researchers further power to detect AMR in newly sequenced species because it focuses on their protein structure rather than just their genomic sequence.7 Since the software is built off of machine learning principles, the development and addition of new verified information and data will make it more accurate over time as the WSU team continues to train the algorithm.7

In the United States alone, such microbes cause more than 2.8 million infections and ailments each year.7 This user-oriented, easy to use program has great potential in aiding scientists in their quest to find and study drug-resistant bacteria with more accuracy and less manual labor. In due time, the world may just see another Fleming-esque revelation, ushering in yet another age of scientific inquiry and discovery.


  1. Arsphenamine – an overview | ScienceDirect Topics. (2020). Retrieved 19 July 2020, from
  2. Chowdhury, A., Call, D., & Broschat, S. (2020). PARGT: a software tool for predicting antimicrobial resistance in bacteria. Scientific Reports, 10(1).
  3. Conniff, R. (2020). How World War II put penicillin into every pharmacy. Retrieved 19 July 2020, from
  4. Fleming, A. (1945). Penicillin: Fleming Nobel Lecture. Retrieved 19 July 2020, from
  5. Machine Learning Algorithms | Microsoft Azure. (2020). Retrieved 22 July 2020, from
  6. Superbugs: What they are, evolution, and what to do. (2020). Retrieved 20 July 2020, from
  7. Washington State University. (2020, July 6). Researchers develop software to find drug-resistant bacteria. ScienceDaily. Retrieved July 19, 2020 from


Bookmark the permalink.

Leave a Reply

Your email address will not be published.