Figure 1.
Topology structures of αTMPs and βTMPs.
(a) Sketches of native TMPs located in biological membrane, where the left one represents an αTMP, and right one is a βTMP. (b) Linear topology of the two TMPs. TM segments are labeled as TMH or TMB respectively according to their TMP types. Orientations of TM segments are described using arrows.
Figure 2.
Alignment accuracy by using topology structure or secondary structure.
The topology structure improves the alignment accuracy of TMFRα (TMFR for αTMPs) comparing with secondary structure, where CNTOP, TMHMM, MEMSAT3 and MEMSAT-SVM were used to general topology structure features, and PSIPRED was for secondary structure feature. TMFRα derived the best alignment accuracy by using CNTOP, which produced more accurate topology structure prediction than other predictors.
Table 1.
Average alignment accuracy of TMFR compared to HHalign.
Figure 3.
Examples showing the correlation of raw score and structure similarity between target and template.
The example of αTMP 1NEK_D is shown in (a), and that of βTMP in (b). Each point on the diagram represents an aligned template. The horizontal axis represents aligned raw score, and the vertical axis shows the corresponding TM-Score. The curve on the diagram is the trend line of data points. The Pearson Correlation Coefficient of 1NEK_D is −0.8120, and that of 1E54_A is −0.8350. Structure similarity is represented using TM-Score. The raw scores generated by TMFR were observed negatively correlating to structure similarities of templates aligned to corresponding target. The templates that have the most similar structures with target are labeled using the PDB classification.
Figure 4.
Correlation of raw score and structure similarity in complete testing dataset.
The Pearson’s correlation coefficients of αTMP and βTMP samples are separately counted in the boxplot above.
Figure 5.
Topological arrangements of top-ranked templates for target 1NEK_D.
1YQ3_D and 1KF6_D are the top-2 templates ranked by raw score.
Table 2.
Comparison of fold recognition performances between OMPs and HHsearch.