Segregating Confident Predictions of Chemicals’ Properties for Virtual Screening of Drugs
In this paper we present a methodology for evaluating the confidence in the prediction of a physicochemical or biological property. Identifying unreliable compounds’ predictions is crucial for the modern drug discovery process.This task is accomplished by the combination of the method of prediction with a self-organizing map. In this way, the method is able to segregate unconfident predictions as well as confident predictions. We applied the method to four different data sets, and we obtained significant differences in the average predictions of our segregation. This approach constitutes a novel way for evaluating confidence, since it not only looks for extrapolation situations but also it identifies interpolation problems.
KeywordsDrug Discovery Applicability Domain Unsupervised Learning Supervised Learning
Unable to display preview. Download preview PDF.
- 2.Barril, X., Hubbard, R.E., Morley, S.D.: Virtual Screening in Structure-Based Drug Discovery. Mini Rev. Med. Chem. 4, 779–791 (2004)Google Scholar
- 8.Jaworska, J., Nikolova-Jeliazkova, N., Aldenberg, T.: QSAR Applicabilty Domain Estimation by Projection of the Training Set Descriptor Space: A Review. Altern. Lab. Anim. 33, 445–459 (2005)Google Scholar
- 10.Yaffe, D., Cohen, Y., Espinosa, G., Arenas, A., Giralt, F.: Fuzzy ARTMAP and Back-Propagation Neural Networks Based Quantitative Structure - Property Relationships (QSPRs) for Octanol-Water Partition Coefficient of Organic Compounds. J. Chem. Inf. Comput. Sci. 42, 162–183 (2002)CrossRefGoogle Scholar
- 11.The Physical Properties Database (PHYSPROP) is marketed by Syracuse Research Corporation (SRC), http://www.syrres.com/esc/