A priori estimation of DNA duplex stability: A Weighted Nearest Neighbor method

Tod Pascal, California Institute of Technology

Quantitative measures of DNA duplex stabilities have been attempted in order to explain everything from binding in protein-DNA complexes to DNA melting. Future applications such as DNA computers, DNA based nano-sensing arrays, and DNA as a component in molecular electronics demand a fundamental and quantitative understanding of DNA structure and the factors affecting it. To this end, a number of experiments have been performed, the most famous results of which were pioneered by the SantaLucia group. Using NMR spectroscopy, they determined the sequence and solution dependence of the DNA double helix; the data were shown to closely fit the Nearest-Neighbor (NN) interaction model. This model assumes that the stability of a DNA duplex depends on the identity and orientation of neighboring base pairs. This approximation therefore reduces the free energy of a base-pair to its interaction with the neighboring bases. Ten different nearest-neighbor interactions are therefore possible in any Watson-Crick DNA duplex structure.



We present the first-ever theoretical study testing the nearest-neighbor model purely from fully atomistic simulations. Using a set of 13 DNA double helices, of random sequence, and variable length, the 10 NN parameters were determined by singular value decomposition (SVD) of the per-atom enthalpies of the entire set. We observe the same trends as the SantaLucia experimental enthalpies, namely: GC > CG > GT ~ AA > GA > CA > GG > AT > TA > CT. Further, there is a >99% correlation between the predicted helical enthalpies from our parameters and that of SantaLucia. More importantly, our parameters allow us to give a priori estimations of the simulation enthalpy of a random sequence of B-DNA to a confidence level of 0.5%.

Abstract Author(s): Tod A. Pascal, Dr. William A Goddard III, Dr. Nagarajan Vaidehi