Meeting the 'systems' challenges in drug discovery: Analytically Integrable Bayesian Methods.

Lead Research Organisation: King's College London
Department Name: Mathematics

Abstract

The statistical methods commonly used in medicine have not kept pace with experimental advances. This leads to inaccurate clinical predictions, poor patient stratification, unnecessary failures of medical trials, and missed clinical opportunities. In this project we develop Bayesian methods for discriminant analysis in which the relevant high-dimensional parameter integrals can be done analytically as opposed to numerically, so that such methods can be applied systematically to high-dimensional (e.g. genomic) medical data. The new methodology will be developed at King's College London, and be applied to ongoing research problems in systems pharmacology at GSK.

Publications

10 25 50
publication icon
Sheikh M (2019) Analysis of overfitting in the regularized Cox model in Journal of Physics A: Mathematical and Theoretical

Studentship Projects

Project Reference Relationship Related To Start End Student Name
BB/N50368X/1 04/01/2016 03/01/2020
1668568 Studentship BB/N50368X/1 04/01/2016 31/12/2019 Mansoor Sheikh
 
Description We extend the standard Bayesian multivariate Gaussian generative data classifier by considering a generalization of the conjugate, normal-Wishart prior distribution and by deriving the hyperparameters analytically via evidence maximization. The behaviour of the optimal hyperparameters is explored in the high-dimensional data regime. The classification accuracy of the resulting generalized model is competitive with state-of-the art Bayesian discriminant analysis methods, but without the usual computational burden of cross-validation.
Exploitation Route The above work is a general classifier which can handle real-valued, high-dimensional data.
Sectors Pharmaceuticals and Medical Biotechnology

URL https://arxiv.org/abs/1712.09813