Table 5.

The origin of the highest scoring model for each of the target sequences is analyzed

Method	Rank 1	Rank 2	Rank 3	Rank 4 to 10
NN-score	82%	10%	3%	5%
NN-all	48%	12%	7%	33%
NN-combined	42%	11%	8%	39%
NN-noscore	26%	15%	10%	50%
NN-model	57%	14%	5%	25%

In the column "rank," all models which were ranked first a prediction server is shown, in "rank 2" second highest ranked etc. Even in NN-score there are some non-first ranked models that are predicted to be better than the first ranked models; these are likely due to noise in the network and the existence of several models with almost identical scores for some targets.