EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3386. A Natural Hermite dissimilarity index randomization for the design of pre-clinical studies, based on a genetic algorithm.

Invited abstract in session TC-40: Data mining biomedical applications, stream Advances in Stochastic Modelling and Learning Methods.

Tuesday, 12:30-14:00
Room: 96 (building: 306)

Authors (first author is the speaker)

1. Sofia Weigle
Johnson and Johnson Pharmaceutical Research

Abstract

Algorithms for randomization of animal studies such as Irini (Weigle et al 2023) are used to allocate the subjects of a study at random into experimental groups. This paper presents a new method that offers an alternative to Irini random allocation a new criterion based on the Natural Hermite index. The idea is that instead of comparing variables across samples individually, the new method compares the multivariate density estimators of the populations. A new method called differential Natural Hermite index was introduced in order to measure the dissimilarity among the multivariate density estimators of the various populations, and it was minimized by applying a genetic algorithm. The genetic algorithm implements a process similar to the Theory of Evolution, for minimizing the differential Natural Hermite dissimilarity for partitions of a dataset into balanced subgroups. Simulations were performed for comparing the new dissimilarity to IRINI and to random allocation via exhaustive search.
The simulations demonstrated the performance superiority of minimizing a fully multivariate dissimilarity such as the Natural Hermite one instead of the combination of univariate dissimilarities implemented by the Irini criterion. The Natural Hermite index that was able to find better partitions than the other two algorithms at a very comparable computational expense.

Keywords

Status: accepted

Back to the list of papers