Conducted a secondary analysis of a multistage, nationally-representative probabilistic sample of NHANES I/NHEFS linked databases examining the role of hematological parameters on subsequent development of cardiovascular disease and cancer. Kaplan-Meier curves were constructed to estimate differential survival among sample population, stratified by each hematological parameter. Cox proportional hazards regression analyses were employed derive hazard ratios associated with each parameter, stratified by gender and ethnicity.
Completed and submitted manuscript proposal for a candidate gene study examining genetic variation in the PI3k/Akt/mTOR pathway and subsequent development of pathological left ventricular hypertrophy in Jackson Heart Study population. Bayesian network analysis (BNA) was employed using R software to curate a concise set of genes within the signaling pathway. Validity and robustness of BNA was ascertained using likelihood ratio tests and bootstrapping for cross-validation. Multi-factor dimensionality reduction (MDR) was employed to finalize a set of tagging SNPs from selected genes for statistical modeling to reduce the effect of multiple testing. Principal components analysis was employed to reduce multicollinearity among covariates in final logistic regression model.
Conducted an exploratory simulation to model the dependency of genetic expression on genetic variants within PI3K/Akt/mTOR signaling pathway to augment the above candidate gene study. Using a priori biological knowledge, dynamic Bayesian network analysis was employed using R and Markov chain Monte Carlo and Python software to account for time-varying probabilistic modeling of RNA expression. Normalized mutual information (MI) between genetic variants and gene expression was calculated to derive statistical models with greatest maximum likelihood. Chi-square tests were then applied to test statistical significance of the final set of MIs.
Employed MedlineR (a library of literature mining algorithms in R), PubMatrix, CoPub, and Bitola to mine academic databases to curate a comprehensive literature on tumor immunology. Methodological algorithms were developed and articles were scored/sorted by scientific strength using SAS Enterprise Miner. A comprehensive literature review was drafted thereafter.
Evaluated the performance of two novel bioassays for the detection of bovine paratuberculosis by employing ROC/AUC analysis. Sensitivity and specificity for each test were calculated, followed by the statistical comparison of each AUC. Estimates and standard errors from each test were utilized to derive a difference score. Difference scores and standard error of the combined tests were then used to calculate z-scores for each respective test. Z-scores were used to obtain p-values to test the null hypothesis that the tests exhibited similar performance.
Employed MedlineR, PubMatrix, and Bitola to mine academic databases to curate a comprehensive literature trained on the association between iron deficiency/iron overload and development of cancer. Using Python, NLP processing was applied to published abstracts to further extract novel associations between the explanatory variables and outcome. Methodological algorithms were developed and articles were scored and sorted by scientific strength using SAS Enterprise Miner. A comprehensive literature review was drafted thereafter.
Following a comprehensive Medline literature review, designed and conducted a meta-analysis using SAS software to examine associations between iron excess and development of Alzheimer’s Disease. Studies were scored/sorted by employing SAS Enterprise Miner. Effect estimates were derived by applying additivity of chi-squared statistics, creating a composite chi-squared test for all studies using study as the effect. Covariates were then added to the model.
Following a comprehensive Medline literature review, designed and conducted a meta-analysis using SAS examining the association between maternal NIDDM and congenital heart defects in neonates. A weighting variable was created to account for strength of study methodology and statistical analysis. Studies were scored and sorted by employing SAS Enterprise Miner. Effect estimates were derived by applying additivity of chi-squared statistics, creating a composite chi-squared test statistic for all studies using study as the effect. Covariates were then added to the model and forest plots were constructed to visually depict study effect.