TransportTP: A two-phase classification approach for membrane transporter prediction and characterization

BMC Bioinformatics

Table 1 The cross-validation results yielded by yeast.

Organism	Num of proteins	Predictions by TransportTP	Annotations in TransportDB	Matches	Mismatches	TransportDB unique	Text mining validated	Recall (%)	Precision (%)	Balanced accuracy (%)
E. coli	5411	589	577	456	6	115	61	79.0	77.4	78.2
A. thaliana	26960	1073	1278	996	1	281	38	77.9	92.8	84.7
O. sativa	56278	1230	1283	1061	0	222	88	82.7	86.3	84.4
C. elegans	20051	906	667	601	1	65	87	90.1	66.3	76.4
D. melanogaster	13890	663	646	535	0	111	26	82.8	80.7	81.7
H. sapiens	37742	1272	1466	1140	3	323	79	77.8	89.6	83.3
Average on model proteomes								81.7	82.2	82.0
P. torridus	1535	165	171	137	1	33	15	80.1	83.0	81.5
P. profundum	5489	550	580	445	4	131	35	76.7	80.9	78.8
D. psychrophila	3234	316	305	242	1	62	38	79.3	76.6	77.9
A. fumigatus	9923	671	619	563	1	55	50	91.0	83.9	87.3
Average on non-model proteomes								81.8	81.1	81.4
Average on all testing proteomes								81.7	81.8	81.8

The proteome of the yeast was used for training and the ten non-yeast proteomes were used for testing at the e-value threshold of 0.1. TransportDB was chosen for the benchmark transporter database. The "matches" column represents the number of proteins predicted by TransportTP and curated by TransportDB with the same TC family or superfamily (the third taxonomic level). The "mismatches" column corresponds to the number of proteins predicted as transporters by both methods but with conflicting TC classification. The column of "TransportDB unique" is the number of proteins annotated by TransportDB but absent in the predictions of TransportTP. The number of "text mining validated" corresponds to the number of proteins not annotated by TransportDB but predicted by TransportTP and validated by our text mining program through the functional annotations together with protein sequences.

ISSN: 1471-2105