TSTMP: Target Selection for human TransMembrane Proteins

Comparison of different methods/databases

Figure 1:
Venn diagram for the results of the different methods.

A comparison was made to similar databases/methods to reveal the extent of the overlap of proteins marked as 'target' between different resources. For this purpose, we selected ModBase/membrane [1] and 3DSpec [2]. Both databases suggest targets for crystallizing and provide a list of proteins with corresponding PDB structure (if any). Since these databases were not updated since 2013 and 2011 (respectively), for a fair comparison we have used the hypothetical '2011' version of TSTMP (i.e. we used the same pipeline that were used to create the recent TSTMP database with the 2011 versions of the source databases). Only proteins listed in all three databases were taken into account.

3DSpecs categorizes proteins as 'Solved', 'Not Solved and Template Found' and 'Not Solved and No Template' which corresponds to our '3D', 'modelable' and 'target' definition. They further categorize proteins by their type, for this comparison IMPs (integral membrane proteins) were selected.

ModBase/membrane does not have such evidence level, therefore we simply used a 25% sequence identity threshold (a transmembrane protein was assumed to be solved or modeled above 25% identity with the assigned PDB entry). To overcome the differences caused by short crystallized fragments, only those proteins were considered as 'modelable'/'3D', where all transmembrane helices were covered by the corresponding PDB entry. The Venn diagram (Figure 1) shows how many of the proteins were listed as 3D/modelable by these resources and the extent of the overlap between them. Outside the circles, the number of the proteins marked as '3D'/'modelable' by all databases are shown. 3DSpecs highly overestimates 3D/modelable proteins, and the agreement between the three methods is only 14%.

1. Pieper U, Schlessinger A, Kloppmann E, Chang GA, Chou JJ, Dumont ME, Fox BG, Fromme P, Hendrickson WA, Malkowski MG, Rees DC, Stokes DL, Stowell MH, Wiener MC, Rost B, Stroud RM, Stevens RC, Sali A. (2013) Coordinating the impact of structural genomics on the human α-helical transmembrane proteome. Nat Struct Mol Biol. 20 135-8. [Medline]
2. Bray JE (2012) Target selection for structural genomics based on combining fold recognition and crystallisation prediction methods: application to the human proteome. J Struct Funct Genomics. 13 37-46. [Medline]