This paper evaluates the results of a protein structure prediction con
test. The predictions were made using threading procedures, which empl
oy techniques for aligning sequences with 3D structures to select the
correct fold of a given sequence from a set of alternatives. Nine diff
erent teams submitted 86 predictions, on a total of 21 target proteins
with little or no sequence homology to proteins of known structure, T
he 3D structures of these proteins were newly determined by experiment
al methods, but not yet published or otherwise available to the predic
tors. The predictions, made from the amino acid sequence alone, thus r
epresent a genuine test of the current performance of threading method
s. Only a subset of all. the predictions is evaluated here. It corresp
onds to the 44 predictions submitted for the 11 target proteins seen t
o adopt known folds. The predictions for the remaining 10 proteins wer
e not analyzed, although weak similarities with known folds may also e
xist in these proteins. We find that threading methods are capable of
identifying the correct fold in many cases, but not reliably enough as
yet. Every team predicts correctly a different set of targets, with v
irtually all targets predicted correctly by at least one team. Also, c
ommon folds such as TIM barrels are recognized more readily than folds
with only a few known examples. However, quite surprisingly, the qual
ity of the sequence-structure alignments, corresponding to correctly r
ecognized folds, is generally very poor, as judged by comparison with
the corresponding 3D structure alignments. Thus, threading can present
ly not be relied upon to derive a detailed 3D model from the amino aci
d sequence. This raises a very intriguing question: how is fold recogn
ition achieved? Our analysis suggests that it may be achieved because
threading procedures maximize hydrophobic interactions in the protein
core, and are reasonably good at recognizing local secondary structure
. (C) 1995 Wiley-Liss, Inc.