Read Time: 7 minutes

Using Computers to Identify Deadly Bacteria Japanese researchers develop a machine learning tool to identify deadly bacteria species based on their unique growth characteristics.

Various industries rely on accurate detection and identification of bacterial species. For example, food-processing plants often have to test their products for bacterial contamination to ensure it is safe for human consumption. In hospitals, a quick and accurate method to identify bacteria can help doctors prescribe the right antibiotics.

Staphylococcus aureus is a bacterium that causes major problems in both the food and medical industries. The bacteria can contaminate food and cause severe diarrhoea. In hospitals, this bacteria causes deadly bloodstream infections in patients. Other relatives of Staphylococcus aureus can also cause infections but are much less severe than S. aureus. It is therefore important that microbiologist can swiftly and accurately identify S. aureus species.

Many of the current techniques that are used to differentiate Staphylococcus species is time consuming and laborious. The existing biochemical tests take a few days to complete. This is especially a problem in hospital settings where delays could be detrimental to patient health.

Newer and more rapid methods are available. For example, mass-spectrometry uses the unique proteins that are made by each species for identification. The downside of this technique is the need for specialized and often expensive equipment.  

To address these issues, a team of scientists from Japan developed a cheap and fast method called colony fingerprinting for distinguishing S. aureus from four other Staphylococcus species. The method is based on the idea that different bacterial species form distinct patterns when growing into structures called colonies on plates in the laboratory. Scientists can then use these unique patterns to identify the Staphylococcus species, just like how the police identify suspects by using fingerprints collected from crime scenes.

To capture images of these colonies, the team created a miniature incubation chamber that holds a plate containing food for the bacteria to grow. They then used a sensitive lens-less camera and a blue LED light for illumination to photograph the growing bacteria. The researchers note that both the camera and light source are relatively inexpensive and easy to obtain. This means that this method can be easily implemented in current  microbiology laboratories.

Once images of the different bacteria species are obtained, the team then had to devise a way to quantify the differences in the colonies. Using a computer program, the researchers came up with 14 ways to measure different characteristics of the bacteria. For example, how long it took the bacteria to grow, how dense the growth was and the colony shapes for each Staphylococcus species.

The team had to then design an automated way to relate those measurements to specific bacterial species, creating a fingerprint. They achieved this automated identification by using machine learning. Machine learning is commonly used to ‘learn’ from data to identify patterns, like how Youtube gives you suggested videos based on your watch list. In the same way, the team taught their computer to recognize patterns in the fingerprints for each bacteria species. Using these patterns, the computer can then make suggestions about the identity of a bacteria species in any sample based on the colony measurements.

Using machine learning, the team was able to achieve 100% accuracy when identifying the different bacterial species. Their machine learning model could also  identify S. aureus colonies when the researchers mixed S. aureus with other unrelated bacteria such as Pseudomonas in the same sample. This means that the model would be useful for identifying bacteria in ‘real-world’ samples from hospitals or food processing plants which would contain mixtures of different bacteria species.

The team believes that implementing this colony fingerprinting technique could give an edge to microbiology laboratories to quickly and accurately identify pathogenic Staphylococcus species. However, the team believes that their fingerprinting technique could benefit from further testing. For example, the colonies in this study were well spaced apart and not growing close together which might not always be the case in real-life samples. In the future, the team intends to test their fingerprinting method with plates that contain dense overlapping colonies.

Article Information

Edited By: Gina Riggio
Source: Colony Fingerprint-Based Discrimination of Staphylococcus species with Machine Learning Approaches
Publication Date: 24 August 2018

Paper Author(s): Yoshiaki Maeda, Yui Sugiyama, Atsushi Kogiso, Tae-Kyu Lim, Manabu Harada, Tomoko Yoshino, Tadashi Matsunaga and Tsuyoshi Tanaka

Paper Institution(s): Tokyo University of Agriculture and Technology, Malcom Co. Ltd., Waseda University

Featured Image Source: https://pixabay.com/en/agar-breeding-ground-red-60571/