Filtered neighbors threading

We present a knowledge-based threading scoring function that exploits the i nformation about protein structure contained in residue packing/neighbor pr eferences. The proposed algorithm eliminates the stereochemically improbabl e physical contacts for each possible sequence-to-structure alignment, We u se this algorithm to "filter" the score of the sequence-to-structure alignm ent. Filtering is dynamic, in the sense that the set of neighbor pairs cont ributing to the alignment score varies during threading. Whether or not a n eighbor pair contributes to the score depends on the threaded amino acids. We use a detailed structure description that encodes amino acid side-chain rotamer and physical contact preferences but does not imprint the fold mode l with the native sequence or native physical contacts. We discretize this description to collect accurate statistics for the scoring function generat ion. We use the original detailed description for the neighbor filtering. O n average, the filtered neighbors threading (FNT) method predicts the seque nce-to-structure alignment twice as accurately as does the "standard" unfil tered neighbors threading. For the set of threadings tested by the PHDthrea der method, the FNT gives predictions with a sequence-to-structure alignmen t accuracy of 46.9%, which amounts to a 74% improvement in alignment sensit ivity compared with PHDthreader predictions. These results show that reduct ion of noise from the observed neighbor pair preferences by filtering leads to noticeable improvements in the predicted sequence-to-structure alignmen ts. Proteins 1999;37:346-359. (C) 1999 Wiley-Liss, Inc.