Protein Structure and Function: Application of Bioinformatics Methods - John Rigden 2014

Fold Recognition
Remote Homology Detection Without Gaps
Heuristic Alignment Rules

The threading approach was originally developed to address Structure/149.html">The problem of assessing sequence-structure compatibility. The number of natural protein folds is finite, which suggests that—given an appropriate energy function and alignment algorithm—threading Methods can succeed where sequence-based methods fail. While sequence-based methods require a detectable level of Homology between the target sequence and a known structure, threading techniques do not.

In the early days of database searching for potential sequence homologs, BLAST and similar approaches predominated. These methods were based on a general scoring function, such as the BLOSUM or PAM Matrices, which reflect the probability of mutational substitution between amino acid residues based on reliably aligned blocks of similar protein sequences. These were straightforward 20x20 lookup tables containing scores for aligning any pair of Amino Acids. Consequently, aligning a hydrophobic residue against another hydrophobic residue (e.g., leucine against valine) received a favorable score, whereas aligning dissimilar residues (e.g., glutamate against Tryptophan) received an unfavorable score. Combining such a scoring function with a standard Dynamic Programming Algorithm yielded relatively modest performance in detecting homologous relationships. However, by searching sequence Databases of known structures and subsequently building a model from the resulting alignment, one could implement one of the simplest methods for Cell/13.html">Protein Structure Prediction (Fig. 2.4a).

An obvious limitation of this approach is that simple 20x20 scoring Functions can only successfully identify close homologs (sharing over 30% sequence identity). Below this threshold, sequences are known to diverge significantly while their three-dimensional structures remain highly similar. As a result, this approach misses numerous homologous relationships whose detection could substantially improve the quality of protein structure prediction.