We are proud to announce that our latest research has been published in the European Medical Journal.
Congratulations to Fredrick Mutisya and the Smartbiotic scientific team on this achievement.
Here is a quick overview by Fredrick Mutisya on how to explore the use of AI against antimicrobial resistance.
Not all missing data are actually “missing.” Sometimes they’re just waiting for a microbiologist to explain them.
One of the biggest challenges in antimicrobial susceptibility (AST) datasets is that the gaps are rarely random. They’re often a consequence of laboratory testing practices and microbiological rules—not simply incomplete data.
In our latest work, we asked a simple question:
Should AI guess first, or should domain knowledge lead the way?
The answer surprised us less than it should have:
- Domain-aware rules achieved excellent specificity and strong overall performance.
- Machine learning improved substantially only after addressing class imbalance.
- The best approach wasn’t AI versus domain knowledge—it was AI built on top of domain expertise.
The takeaway is one I’ll keep repeating:
In healthcare AI, biology should constrain the algorithms—not the other way around.
As we build clinical AI systems, success won’t come from bigger models alone. It will come from combining expert knowledge, deterministic clinical rules, and machine learning where it adds the most value.
Proud to see this work published and grateful to my co-authors and collaborators at Smartbiotic.
Access the publication here :
https://www.emjreviews.com/microbiology-infectious-diseases/abstract/domain-aware-versus-machine-learning-imputation-for-sparse-antimicrobial-susceptibility-data-j10126/
