011: Analysis of Supervised Machine Learning Algorithms & Computational Validation of the Single Nucleotide Polymorphisms of Antioxidant Enzymes & Oxidative Stress Biomarkers in Preterm Neonates

Statement of Purpose: Oxidative stress play a significant pathophysiological role in neonatal respiratory distress syndrome (RDS). Similarly, oxidative stress predisposes to significant alterations in liver functions. Hence, we carried out the present study evaluating various machine learning algorithms (MLAs) for identifying significant oxidative stress and genetic biomarkers for RDS and significant alterations in liver functions.

Description of Methods & Materials: Preterm neonates were recruited, and oxidative stress biomarkers such as malondialdehyde, 8-hydroxy-2-deoxy guanosine, total antioxidant capacity and advanced oxidation end products (AOPP) were estimated. Single nucleotide polymorphisms (SNPs) in catalase (CAT), superoxide dismutase (SOD1 and SOD2), glutathione peroxidase (GPX) and glutathione S-transferases were estimated. MLAs [logistic regression, discriminant analysis, artificial neural network (ANN), Bayesian network, C5.0 algorithm, classification and regression tree analysis (CART) and chi-square automatic interaction detector (CHAID)] were evaluated independently for predicting the significant alterations in liver functions and RDS. The area under the curve (AUC) was used to measure predictive accuracy. Bioinformatic tools were used to identify the protein-protein interaction analysis between the significant SNPs.

Data & Results: The C5.0 algorithm was observed with the best predictive accuracy (AUC = 0.63) for identifying significant changes in liver functions where various oxidative biomarkers and SNPs in the CAT were identified as the most important predictors. For RDS, the Bayesian network had the best predictive accuracy (AUC = 0.6) for oxidative biomarkers that have identified AOPP concentrations in the peripheral blood, with logistic regression analysis as the best model (AUC = 0.59) for identifying genetic markers of which eNOS 1 was identified as the most critical predictor variable. Bioinformatic tools have identified the SOD2 enzyme to have significant interaction with other genes in the family, such as SOD1 and SOD3. Further, CAT, SOD2 and GPX3 have strong interaction than NOS3.

Interpretation, Conclusion or Significance: We have shown that MLAs carry great potential in identifying the potential genetic and oxidative stress biomarkers in neonatal RDS and significant changes in liver functions. Bioinformatics tools validate the altered enzymatic activity of the SNPs evaluated in the present study. Validation in prospective studies is needed urgently.

Disclosures: