Prediction by Nonparametric Posterior Estimation in Virtual Screening

Download

PDF (Poster) - Published Version Download (116Kb)

Description/Abstract

The ability to rank molecules according to their effectiveness in some domain, e.g. pesticide, drug, is important owing to the cost of synthesising and testing chemical compounds. Virtual screening seeks to do this computationally with potential savings of millions of pounds and large profits associated with reduced time to market. Recently, binary kernel discrimination (BKD) is introduced and becoming popular in Chemoinformatics domain. It produces scores based on the estimated likelihood ratio of active to inactive compounds that are then ranked. The likelihoods are estimated through a Parzen Windows approach using the binomial distribution function (to accommodate binary descriptor or "fingerprint" vectors representing the presence, or not, of certain sub-structural arrangements of atoms) in place of the usual Gaussian choice. This research aims to compute the likelihood ratio via direct estimate of posterior probability by using non-parametric generalisation of logistic regression the so-called “Kernel Logistic Regression”. Furthermore, complexity is then controlled by penalising the likelihood function by Lq-norm. The compounds are then rank descending on the basis of posterior probability. The 11 activity classes from the MDL Drug Data Report (MDDR) database are used. The results are found to be less accurate than a currently leading approach but are still comparable in a number of cases.