Filtering of false positive microRNA candidates by a clustering-based approach

BMC Bioinformatics

Table 6 Results of our clustering-based approach when applied on clusters with more than one TP.

	Human		Mouse (All)		Mouse (Distinct)		Rat
Before/After filtering	before	after	before	after	before	after	before	after
# of predictions	620	335	632	363	616	350	525	224
# of TPs	206	189	195	168	179	155	121	99
# of real miRNAs	220	220	234	234	202	202	132	132
SE	93.64%	85.91%	83.33%	71.79%	88.61%	76.73%	91.67%	75.00%
Change in SE	-	-7.73%	-	-11.54%	-	-11.88%	-	-16.67%
PPV	33.23%	56.42%	30.85%	46.28%	29.06%	44.29%	23.05%	44.20%
Change in PPV	-	+23.19%	-	+15.43%	-	+15.23%	-	+21.15%

If we exclude the clusters which bear no TPs or just one TP among the candidates predicted by ProMirII-g, we can see a great improvement in PPV without a significant effect on SE after our filtering approach is applied. The results agree with the principle of our approach, which is developed based on the phenomenon of miRNA clustering. In other words, if there are no clustered miRNAs in a sequence, our approach is not going to work properly. This table presents the results of a fairer comparison, suggesting that our approach is effective in filtering FPs.

ISSN: 1471-2105