How many dimensions are required to approximate the potential energy landscape of a model protein?

A scheme to approximate the multidimensional potential energy landscape in terms of a minimal number of degrees of freedom is proposed using a linear transformation of the original atomic Cartesian coordinates. For one particular off-lattice model protein the inherent frustration can only be reproduced satisfactorily when a relatively large number of coordinates are employed. However, when this frustration is removed in a Go-type model, the number of coordinates required is significantly lower, especially around the global potential energy minimum. To aid our interpretation of the results we consider modified disconnectivity graphs where a measure of the structural diversity and a metric relation between the stationary points are incorporated.