Threading algorithm

TR Defay, EE Cohen. Multiple sequence information for threading algorithms. J Mol Biol 262 314-323, 1996. 

A major problem in predicting protein structure is the computational intractability. A short, 100-residue protein will contain at least 100 side-chain-to-side-chain or side-chain-to-solvent interactions. The orientation of each of these interactions will lead to cascading effects throughout the protein. Comparative modeling, threading algorithms, and de novo predictions seek to predict protein structure in reasonable execution times. 

The ability to recognize the way in which a protein sequence is folded in three dimensions should enable us to model the interactions of specific side-chains in a manner that is simply not possible when considering proteins entirely at the sequence level. This notion has resulted in sequence threading algorithms that assess the level of compatibility of a sequence with a database of fold patterns (65, 66). The principal downside to this approach is that novel structural types cannot be pre- dieted, because at least one example of each fold type must be present in the fold pattern database. Structural genomics may be the means whereby fold pattern databases can be populated with sufficient data to make them useful as predictive tools. 

Besides expressions such as Eq. (5.7) one can construct expressions with reduced variance. These involve the configurational eigenvalue of G or Jtf in the same way this was done in our discussion of these single-thread algorithm. 

The LOOPSEARCH module of the SYBYL package [60] is a threading algorithm used to find good sfarfing conformations for the loops. 

Unfortunately, such methods require the exact placement of atoms within protein side chains and are inapplicable to the inexact, low-resolution predicted structures generated by the state-of-the-art ab initio folding and threading algorithms (see Sections IV-VI). These methods are required when the sequence identity of the sequence of interest to solved structures is too low to use comparative modeling. To address this need, Skolnick and Fetrow have recently developed fuzzy, inexact descriptors of protein functional sites [8]. They are applicable to both high-resolution, experimental structures and low-resolution (backbone RMSD 4—6 A from native) structures. These descriptors are a-carbon-based, fuzzy functional forms (FFFs). Initially, they created FFFs for the disulfide oxidoreductase [8,10] and a/p-hydrolase catalytic active sites [11] (an additional 198 have now been bmlt, with comparable results [234]). 

Figure 10 Schematic of a threading algorithm. A search pattern is slid along a sequence and evaluated in the corresponding structure s conformation. Conformations that make a significant number of favorable interactions are presumed to be native-like...

---