[A novel metabolomic data scaling method based on K-L divergence].

Area of Science:

• Metabolomics
• Bioinformatics
• Analytical Chemistry

Background:

• NMR metabolomics generates complex datasets requiring robust preprocessing.
• Existing scaling methods may not optimally handle non-Gaussian distributions or effectively differentiate biological signals from noise.
• Supervised methods can improve data analysis by incorporating group information.

Purpose of the Study:

• To introduce a new supervised scaling method for NMR metabolomic data based on Kullback-Leibler (K-L) divergence.
• To evaluate the efficacy of K-L scaling in noise reduction, variable weighting, and enhancing multivariate model performance.
• To assess the method's ability to improve interpretability and facilitate the identification of metabolic signatures.

Main Methods:

• Development of K-L scaling, a supervised method incorporating group information.
• Standardization of variables to unit variance followed by variance adjustment using K-L divergence.
• Application of K-L scaling to a human urine H-NMR metabolomic dataset.
• Comparison of K-L scaling with other methods in multivariate analysis (PCR, PLS-DA).

Main Results:

• K-L scaling effectively suppresses noise in multivariate models of NMR metabolomic data.
• The method enhances the weights of biologically relevant variables while reducing the influence of noise.
• Improved interpretability and predictability of Principal Component Regression (PCR) and Partial Least Squares Discriminant Analysis (PLS-DA) models were observed.
• Facilitation of metabolic signature identification was demonstrated.

Conclusions:

• K-L scaling is an efficient supervised method for preprocessing NMR metabolomic data.
• The method offers advantages in noise reduction and highlighting significant biological variables.
• K-L scaling shows potential as a valuable alternative for NMR metabolomic data analysis, improving model performance and interpretability.