Antimicrobial resistance (AMR) is perhaps among the most urgent threats to human health. Understanding how superbugs get their immunity can help us fight back, but there are many different mechanisms involved that make it difficult to find out. The project AMR-XAI aims to uncover resistance mechanisms using AI, by developing new algorithms and working with real-world data from clinical care.
Helmholtz AI project call showcase: Crushing antimicrobial resistance using explainable AI
How can AI models help tackle the persisting spread of antimicrobial resistance? Read in this week’s Helmholtz AI project showcase how researchers at the Helmholtz Centre for Infection Research (HZI) and the Helmholtz Center for Information Security (CISPA) are working jointly on this challenge.
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
We are Prof. Dr. Olga V. Kalinina, bioinformatician from the Helmholtz Institute for Pharmaceutical Research Saarland / Helmholtz Centre for Infection Research (HIPS/HZI); and Prof. Dr. Jilles Vreeken, machine learning specialist from the Helmholtz Center for Information Security (CISPA). Together we are working on the AMR-XAI project: Crushing Antimicrobial Resistance using Explainable AI.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
Since their discovery over a century ago, antibiotics have greatly improved human life expectancy and quality: many diseases went from life-threatening to mild inconveniences. Miss- and over-usage of these drugs, however, has caused microbes to develop resistance to even the most advanced drugs; diseases once considered conquered are becoming devastating again. While individual resistance mutations are well-researched, knowing which new mutations can cause antimicrobial resistance is key to developing drugs that reliably sidestep microbial defences.
In this project we aim to gain this knowledge via explainable artificial intelligence, a type of AI where we can understand how its predictions are being made. With this tool we will develop and apply novel methods for discovering mutation patterns that are relevant to one or more classes of resistance. That is, we propose to learn a small set of easily interpretable models that together explain the resistance mechanisms in the data. For this we will use statistically robust methods to discover significant subgroups, as well as information theoretic approaches to establish succinct sets of rules robust to noise or other nonessential data.
Key to our success is the tight integration of domain expertise into the development of the new algorithms, early evaluation on real-world data, and the potential available in the host institute to evaluate particularly promising results in the lab.
How important has the Helmholtz AI funding and platform been to carry out this project?
Helmholtz AI funding gave us a chance to start this cooperation. Given the high-risk nature of the project and the large gap between the fields of study that it bridges, it would have been next to impossible to have it funded by traditional funding bodies.
The funding through Helmholtz AI provides a unique opportunity to bring expertise on the biological, bioinformatics as well as machine learning side towards developing novel algorithms that work in practice for solving AMR.