projects
Stochastic comparisons:
Demler, O.V. and Demler, I.A., 2023. Non-transitivity of the Win Ratio and Area Under the Receiver Operating Characteristics Curve (AUC): a case for evaluating the strength of stochastic comparisons. arXiv preprint https://arxiv.org/abs/2309.01791.
Below is a visualization of the space of {Pr(x>y), Pr(y>z), Pr(z>x)} generated by any three independent random variables x, y and z. Non-transitive subspace is in yellow color (Trybula, 1961).
Below is a visialization of the space of of triplets of win ratios generated by any three independent random variables x, y and z. Non-transitive subspace is in yellow color (Trybula, 1961), Demler et. al. manuscript coming soon.
*in collaboration with Ilona Demler
Biomarker discovery large omics studies:
Advanced high-resolution LC-MS techniques are able to detect compounds at concentrations as low as nanograms per milliliter. Below is how the composition of blood changes upon the intake of 840 mg/day of fish oil (EPA+DHA). (Demler et al., Metabolites, 2020) & upcoming publications.
*in collaboration with Dr. Samia Mora and members of the Division of Preventive Medicine
Interpretable machine learning:
Many machine learning models operate as black boxes. Below is a heatmap depicting the 10-year risk of cardiovascular disease based on systolic blood pressure and total cholesterol levels for specific age groups, genders, races, type 2 diabetes statuses, and smoking habits. Within the heatmap, the boxes illustrate the risk fluctuations inferred by the black box model. However, it’s possible to constrain the model to adhere to established biological relationships with risk, thereby transforming a black box model into a comprehensively interpretable prognostic model. (The manuscript is currently under peer review)
*in collaboration with Aniket Zinzuwadia, Dr. Samia Mora, Olga Mineeva and members of the Division of Preventive Medicine
Evaluating calibration of survival models (modified Hosmer-Lemeshow test):
(Demler et. al., Statistics in Medicine, 2015, 2017)
*in collaboration with Drs. Nancy Cook and Nina Paynter and members of Division of Preventive Medicine
Evaluating the discriminative ability of binary classifiers and prognostic risk prediction models using AUCROC:
(Demler et. al., Statistics in Medicine, 2012, 2013, 2017)
*in collaboration with Drs. Ralph D’Agostino Sr., Michael Pencina, and other colleagues