Recently in the Washington State University, researchers have created a new method to identify antimicrobial-resistance (AMR). The new method uses game theory to choose the important features to be used in their machine learning model. While there were many attempts in identifying AMR with machine learning, there was never a good strategy to identify the important features to be used.
Most of the strategies used in the past had either resulted in high computational overhead or low accuracy. However researchers found out using game theory they can now include features that are poorly predictive as single variables, which are traditionally rejected. The method uses the behavior of different features and consider how they behave together as a whole. Using the data generated from game theory, they can then train their machine learning model to accurately predict AMR. The resulting algorithm has an accuracy ranging from 93% – 99% as shown in the figure below.
“This novel game theory approach is especially powerful because features are chosen on the basis of how well they work together as a whole to identify likely antimicrobial-resistance genes — taking into account both the relevance and interdependency of features,” – Broschat, one of the researchers in developing this new method of identifying AMR.
Source:
Abu Sayed Chowdhury, Douglas R. Call, Shira L. Broschat. Antimicrobial Resistance Prediction for Gram-Negative Bacteria via Game Theory-Based Feature Evaluation. Scientific Reports, 2019; 9 (1) DOI: 10.1038/s41598-019-50686-z