Deducing protein structures using logic programming: exploiting minimum data of diverse types

The extent to which a protein can be modeled from constraint data depends on the amount and quality of the data. This report quantifies a relationship between the amount of data and the achievable model resolution. In an information-theoretic framework the number of bits of information per residue needed to constrain a solution was calculated. The number of bits provided by different kinds of constraints was estimated from a tetrahedral lattice where all unique molecules of 6, 9, …, 21 atoms were enumerated. Subsets of these molecules consistent with different constraint sets were then chosen, counted, and the root-mean-square distance between them calculated. This provided the desired relations. In a discrete system the number of possible models can be severely limited with relatively few constraints. An expert system that can model a protein from data of different types was built to illustrate the principle and was tested using known proteins as examples. C-α resolutions of 5 Å are obtainable from 5 bits of information per amino acid and, in principle, from data that could be rapidly collected using standard biophysical techniques.