The dense alignment surface (DAS) transmembrane (TM) prediction method was first published
more than 25 years ago. DAS was the one of the earliest tools to discriminate TM proteins
from globular ones and to predict the sequence positions of TM helices in proteins
with high accuracy from their amino acid sequence alone. The algorithmic improvements
that followed in 2002 (DAS-TMfilter) made it one of the best performing tools among
those relying on local sequence information for TM prediction. Since then, many more
experimental data about membrane proteins (including thousands of 3D structures of
membrane proteins) have accumulated but there has been no significant improvement
concerning performance in the area of TM helix prediction tools. Here, we report a
new implementation of the DAS-TMfilter prediction web server. We reevaluated the performance
of the method using a five-times-larger, updated test dataset. We found that the method
performs at essentially the same accuracy as the original even without any change
to the parametrization of the program despite the much larger dataset. Thus, the approach
captures the physico-chemistry of TM helices well, essentially solving this scientific
problem.
