End-To-End Deep Learning Explains Antimicrobial Resistance in Peak-Picking-Free MALDI-MS Data

Area of Science:

• Clinical microbiology
• Computational biology
• Spectrometry data analysis

Background:

• Mass spectrometry is crucial for identifying infectious microbes in clinical settings worldwide.
• Vast amounts of mass spectrometry data can be leveraged with advanced analysis methods for deeper insights.
• Predicting antimicrobial resistance is vital for effective treatment strategies.

Purpose of the Study:

• To develop an end-to-end deep learning model for predicting antibiotic resistance directly from raw MALDI-MS data.
• To bypass conventional data processing steps like peak-picking for a streamlined workflow.
• To demonstrate the potential of deep learning in enhancing MALDI-MS data analysis for clinical applications.

Main Methods:

• Utilized a 1-dimensional convolutional neural network (1D CNN) architecture.
• Applied the model to (almost) raw MALDI-MS data, avoiding traditional peak extraction.
• Trained and validated the model across diverse antimicrobial resistance phenotypes, time points, and locations.

Main Results:

• Achieved state-of-the-art performance with Area Under the Curve (AUC) values ranging from 0.93 to 0.99 for all tested antimicrobial resistance phenotypes.
• Demonstrated robust model validation across different time periods and geographical locations.
• Feature attribution analysis provided insights into model behavior and potential workflow improvements.

Conclusions:

• Reliable antibiotic resistance phenotyping using MALDI-MS data is achievable with advanced deep learning techniques.
• End-to-end deep learning models offer significant advantages for analyzing spectrometry data.
• This approach can improve the efficiency and accuracy of infectious disease diagnostics.