Send to

Choose Destination
Integr Biol (Camb). 2017 Jul 17;9(7):627-641. doi: 10.1039/c7ib00079k.

The maximum penalty criterion for ridge regression: application to the calibration of the force constant in elastic network models.

Author information

Machine Learning Group, Université Libre de Bruxelles (ULB), Boulevard du Triomphe CP 212, 1050 Brussels, Belgium.


Tikhonov regularization, or ridge regression, is a popular technique to deal with collinearity in multivariate regression. We unveil a formal analogy between ridge regression and statistical mechanics, where the objective function is comparable to a free energy, and the ridge parameter plays the role of temperature. This analogy suggests two novel criteria for selecting a suitable ridge parameter: specific-heat (Cv) and maximum penalty (MP). We apply these fits to evaluate the relative contributions of rigid-body and internal fluctuations, which are typically highly collinear, to crystallographic B-factors. This issue is particularly important for computational models of protein dynamics, such as the elastic network model (ENM), since the amplitude of the predicted internal motion is commonly calibrated using B-factor data. After validation on simulated datasets, our results indicate that rigid-body motions account on average for more than 80% of the amplitude of B-factors. Furthermore, we evaluate the ability of different fits to reproduce the amplitudes of internal fluctuations in X-ray ensembles from the B-factors in the corresponding single X-ray structures. The new ridge criteria are shown to be markedly superior to the commonly used two-parameter fit that neglects rigid-body rotations and to the full fits regularized under generalized cross-validation. In conclusion, the proposed fits ensure a more robust calibration of the ENM force constant and should prove valuable in other applications.

[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Royal Society of Chemistry
Loading ...
Support Center