Symmetry Principles in Optimization Problems: an application to Protein Stability Prediction

Katrien Bernaerts*, F. Pucci, M. Rooman, D. Gillis, Dimitri Gilis

*Corresponding author for this work

Research output: Contribution to journalConference article in journalAcademicpeer-review

Abstract

In this paper, we show how the adequate use of the intrinsic symmetry of a system when setting up its model structure can avoid unwanted biases in the parameter optimization phase. The playground of our analysis is the prediction of protein thermodynamic stability changes upon single amino acid substitutions (point mutations). Using a simple artificial neural network (ANN), sixteen different energy-like contributions are combined to predict the change in folding free energy (Delta Delta G). We show that the presence of terms violating the symmetry under inverse mutations induces a bias towards the dataset on which the ANN is trained, even if a strict n-fold cross-validation procedure is performed. A completely symmetric free energy functional is then introduced, which gives predictions that are slightly less efficient in terms of root mean square error with respect to the experimental Delta Delta G's, but appear to be basically independent of the training dataset and are thus more satisfactory. (C) 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Original languageEnglish
Pages (from-to)458-463
Number of pages6
JournalIFAC-PapersOnLine
Volume48
Issue number1
DOIs
Publication statusPublished - 2015
Event8th Vienna International Conference on Mathematical Modelling - Vienna, Austria
Duration: 18 Feb 201520 Feb 2015

Cite this