Imputing missing minimum inhibitory concentration (MIC) values for Pseudomonas aeruginosa strains with a Denoising AutoEncoder
Elsa Denakpo, Nicolas Dias, Johann D D Pitout, Thierry Naas, Dylan R. Pillai, Flora Jay, Bogdan Iuliu Iorga · bioRxiv (Cold Spring Harbor Laboratory) · 2025
Abstract Pseudomonas aeruginosa is a problematic pathogen with complex antibiotic resistance patterns. In clinical practice, minimum inhibitory concentration (MIC) tests typically focus on a limited subset of antibiotics, hindering a comprehensive assessment of a strain’s resistance profile. Here, we introduce MICFiller, a Denoising AutoEncoder (DAE) model designed to impute missing MIC values for 14 antibiotics in Pseudomonas aeruginosa within a specific dilution range by leveraging known MIC measurements for other antibiotics in the same strain. We evaluated the performance of DAE against two other commonly used methods: Multiple Imputation by Chained Equations (MICE) and simple median imputation. The DAE achieved the highest balanced 1-tier accuracy for most antibiotics, with performance closely matching that of MICE. MICFiller is freely accessible through a user-friendly web interface at http://iorgalab.org:4567/micfiller , offering clinicians a more complete view of a strain’s antibiotic resistance profile.