Dipeptide Preferences and Molecular Structure

.....The ability to correlate the amino acid sequence of a protein with higher order molecular structure has a been a goal of experimental and theoretical protein chemists for decades. A number of statistical approaches have been explored with some success. The original probabilistic method set forth in studies by Chou and Fasman are a landmark in this field, despite the rather limited success of the original approach to actually predict native protein structure. Subsequent studies have built upon this probabilistic approach employing a variety of modifications. In general, the limitations of Chou and Fasman's probabilistic methodology have been attributed to the difficulty in establishing accurate statistics correlating the occurrence of a particular amino acid within an established protein conformation.
.....At DanPatGenomics we maintain that the established values relating the probability of finding a given amino acid within a particular protein conformation are essentially accurate. We suggest instead that the difficulty in predicting protein structure from a linear string of amino acids lies in the mathmatical treatment of the values used to calculate elements of the classic three-state (alpha, beta, turn) protein structure elements, not in the assignment of individual correlation values.
.....Our computational tool, 2DSEQ.TAB, for sequence-based structure prediction exploits our previous experience in analyzing proteins at the two-amino-acid level and employs a novel series of algorithms which calculate the likelihood of finding alpha, beta, turn or random coil structures based on the linear arrangement of dipeptide sequence elements. This approach has proven to be superior in predicting higher order protein structure than other probabilistic methods.