XModNN: Explainable Modular Neural Network for Identifying Clinical Parameters and Disease Biomarkers
Author Information
Author(s): Jan Oldenburg, Jonas Wagner, Sascha Troschke-Meurer, Jessica Plietz, Lars Kaderali, Henry Völzke, Matthias Nauck, Georg Homuth, Uwe Völker, Stefan Simm
Primary Institution: University Medicine Greifswald
Hypothesis
Can the Explainable Modular Neural Network (XModNN) effectively identify biomarkers and classify diseases using transcriptomic datasets?
Conclusion
XModNN successfully identifies biomarkers and classifies diseases with fewer candidate biomarkers compared to standard statistical approaches.
Supporting Evidence
- XModNN reduced the number of important genes by a factor of 2 to 35 compared to other models.
- XModNN achieved a median balanced accuracy of 1.00 in sex classification.
- XModNN identified 26 genes and six pathways crucial for neuroblastoma cell state discrimination.
- XModNN's architecture allows training on limited examples while maintaining performance.
- XModNN integrates biological insights into its design, enhancing interpretability.
Takeaway
XModNN is a smart computer program that helps doctors find important clues about diseases by looking at tiny bits of information from our genes.
Methodology
XModNN uses a modular neural network architecture with weighted multi-loss progressive training and layer-wise relevance propagation for feature selection and explanation.
Potential Biases
Potential biases may arise from the limited number of samples and the complexity of molecular interactions.
Limitations
The study may be limited by the small sample sizes in some use cases and the complexity of the models used.
Participant Demographics
Participants aged 20 to 79 years, with 434 males and 555 females in the sex discrimination study.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website