skip to main content

Title: Not All Features Are Equal: Discovering Essential Features for Preserving Prediction Privacy
Authors:
; ; ; ; ;
Award ID(s):
1703812
Publication Date:
NSF-PAR ID:
10294357
Journal Name:
International Web Conference
Page Range or eLocation-ID:
669 to 680
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract Aims We sought to investigate whether artificial intelligence (AI) and specifically deep neural networks (NNs) for electrocardiogram (ECG) signal analysis can be explained using human-selected features. We also sought to quantify such explainability and test if the AI model learns features that are similar to a human expert. Methods and results We used a set of 100 000 ECGs that were annotated by human explainable features. We applied both linear and non-linear models to predict published ECG AI models output for the detection of patients’ age and sex. We further used canonical correlation analysis to quantify the amount of shared information between the NN features and human-selected features. We reconstructed single human-selected ECG features from the unexplained NN features using a simple linear model. We noticed a strong correlation between the simple models and the AI output (R2 of 0.49–0.57 for the linear models and R2 of 0.69–0.70 for the non-linear models). We found that the correlation of the human explainable features with either 13 of the strongest age AI features or 15 of the strongest sex AI features was above 0.85 (for comparison, the first 14 principal components explain 90% of the human feature variance). We linearly reconstructedmore »single human-selected ECG features from the AI features with R2 up to 0.86. Conclusion This work shows that NNs for ECG signals extract features in a similar manner to human experts and that they also generate additional novel features that help achieve superior performance.« less
  2. The increased interest in sequencing cyanobacterial genomes has allowed the identification of new homologs to both the N-terminal domain (NTD) and C-terminal domain (CTD) of the Orange Carotenoid Protein (OCP). The N-terminal domain homologs are known as Helical Carotenoid Proteins (HCPs). Although some of these paralogs have been reported to act as singlet oxygen quenchers, their distinct functional roles remain unclear. One of these paralogs (HCP2) exclusively binds canthaxanthin (CAN) and its crystal structure has been recently characterized. Its absorption spectrum is significantly red-shifted, in comparison to the protein in solution, due to a dimerization where the two carotenoids are closely placed, favoring an electronic coupling interaction. Both the crystal and solution spectra are red-shifted by more than 50 nm when compared to canthaxanthin in solution. Using molecular dynamics (MD) and quantum mechanical/molecular mechanical (QM/MM) studies of HCP2, we aim to simulate these shifts as well as obtain insight into the environmental and coupling effects of carotenoid–protein interactions.