|
|
|
|
|
|
|
|
| About Olga Obrezanova (BioFocus DPI) |
|
Dr. Olga Obrezanova graduated from Rostov University in Russia and obtained her Ph.D. in Applied Mathematics from the same university in 1995. She then worked for four years as an Assistant Professor at Rostov University researching underwater acoustics. Between 1999 and 2003, Olga worked as a Research Associate at the University of Cambridge. Her research involved investigation into crack propagation in solid materials and fracture mechanics.
In 2005 Olga joined Inpharmatica, a drug discovery company that merged with BioFocus DPI at the end of 2006. Her work has been focused on applying statistical modeling and machine learning methods to problems in drug discovery. In particular, Olga has been developing and using new computational techniques to build QSAR models of ADME properties. Most recently, Olga’s key research has been around algorithms enabling automatic model generation. Olga has been working as part of a team developing a commercial, decision making, software platform for compound design, optimization and prioritization (StarDrop) where she has led the research and design process for the “Auto-Modeler” tool. She is based in Cambridge, UK.
|
|
Automatic QSAR Modeling of Blood-Brain Barrier Penetration by Gaussian Processes Method
Olga Obrezanova*, Joelle M.R. Gola, Edmund J. Champness, Matthew D. Segall, BioFocus DPI, Chesterford Park, Saffron Walden, CB10 1XL, UK
Blood-brain barrier (BBB) penetration is often one of the key properties considered during ADME (Absorption, Distribution, Metabolism and Excretion) studies in drug discovery. There have been a large number of in silico approaches to modeling BBB penetration reported in the literature describing either continuous models predicting the logarithm of the brain-blood concentration ratio (logBB) or classification models predicting BBB+/-. In this presentation we will discuss the results of modeling blood-brain barrier penetration by employing an automatic model generation process based on Gaussian Processes, a computational, machine learning technique.
The rapid design-test-redesign cycles of modern drug discovery and the demand for fast model (re)building whenever data becomes available have given rise to a trend to develop computational algorithms for automatic model generation. Automatic modeling processes allow computational scientists to explore large numbers of modeling approaches very efficiently and make QSAR/QSPR model building accessible to non-experts. Automatic model generation requires unsupervised, computational techniques that are not dependant on any input from a user, are able to deal with a large number of descriptors and are not prone to overtraining. The automatic model generation process which we will present is based on Gaussian Processes, a powerful non-linear probabilistic method. The Gaussian Processes technique is highly appropriate for automatic model generation; it does not require subjective determination of model parameters, it is able to handle a large pool of descriptors and select the important ones, it is inherently resistant to overtraining and offers a way of estimating the uncertainty in predictions. In our previous work [1] we developed new techniques for implementing the Gaussian Processes method for modeling continuous data and compared their performance with other modeling techniques. In this presentation we will demonstrate how we have extended our Gaussian Processes techniques to model categorical data.
We will present an automatic model generation process for building QSAR models and describe the stages of the process that ensure models are built and validated within a rigorous framework; descriptor calculation, splitting data into training, validation and test sets, descriptor filtering, application of modeling techniques and selection of the best model. We will then demonstrate the application of the automatic model generation process to modeling two blood-brain barrier penetration data sets; a continuous logBB data set and a classified BBB+/- data set. We will present examples of automatically generating both continuous and classification models, compare the resulting models with ‘manually’ built models and finally demonstrate the results of rebuilding an existing model by including new data points.
Reference:
1. Obrezanova, O.; Csányi, G.; Gola, J.M.R.; Segall, M.D. Gaussian Processes: A Method for Automatic QSAR Modeling of ADME Properties. J. Chem. Inf. Model. 2007, 47, 1847-1857.
|
|
|
|
|
|
|
|
|
|
|
