The first step in the process of obtaining bond, angle, and dihedral parameters was to establish the mean bond and angle geometries. Bond lengths were taken from Bruce Gelin's Ph. D. thesis (the dipeptide modelling section), and agree quite well with other tabulations of standard bond lengths such as the CRC Handbook and Karplus and Porter. Angles were chosen as the average value observed in protein crystal structures available at better than 2.0 angstroms resolution on the Brookhaven Protein Data Bank tape. This includes the following structures:
These coordinates and their sequences were read off the tape and tables of the angle geometries built using the CHARMM program. The average values for each angle type were collected over all of these structures weighted equally. This compilation allowed the averaging of twenty to fifteen hundred occurrences for each angle type, with variances ranging from less than one degree to as much as fifteen degrees, and ranges of two to sixty five degrees. In general these average values are within two degrees of the PARMFIX10 values. It should be noted that all of the above structures are crystallographically refined. A certain amount of bias is therefore inevitable, but I feel justified by the higher resolutions that these methods achieve.