Table 3. Accuracy of GINDEL trained by combination of different features.

Length	Gntps	D. C.	S. F.	R.	D. S.	D. C. R.	R. S. F.	D. C. S. F.	All
50–75	294	0.8844	0.9218	0.7721	0.9252	0.8844	0.9082	0.9388	0.9592
75–100	294	0.8878	0.9728	0.8844	0.9490	0.9354	0.9728	0.9762	0.9796
100–150	238	0.8866	0.9958	0.8277	0.9454	0.9664	0.9874	1.0000	1.0000
150–200	154	0.9351	0.9610	0.9351	0.9870	0.9870	0.9545	0.9870	0.9870
200–300	224	0.9420	0.9200	0.8259	0.9464	0.9688	0.9375	0.9643	0.9688
300–400	420	0.9262	0.9333	0.8262	0.9381	0.9619	0.9405	0.9643	0.9667
400–500	182	0.9615	0.9670	0.8736	0.9560	1.0000	0.9780	1.0000	1.0000
500–800	140	0.9929	0.9714	0.9214	0.9929	1.0000	0.9714	1.0000	1.0000
800–1100	168	1.0000	0.9464	0.8929	1.0000	1.0000	0.9583	1.0000	1.0000
1100–1500	168	0.9702	0.9643	0.9226	0.9702	1.0000	0.9702	0.9940	1.0000
1500–2000	210	0.9714	0.9762	0.9476	0.9857	1.0000	0.9762	0.9952	0.9952
2000–4000	252	0.9762	0.9444	0.9246	0.9722	0.9960	0.9841	0.9881	0.9960
4000–9000	238	0.9748	0.9500	0.8277	0.9706	0.9748	0.9538	0.9832	0.9874
>9000	112	0.9821	0.9821	0.8660	0.9821	1.0000	0.9821	1.0000	1.0000

Results on the percentage of correctly called genotypes are given for deletions with different lengths. Gntps represents the total number of genotypes in specific length. D., C., S., F., and R. represents # of discordant read-pairs, # of concordant read-pairs, # of split-mapped reads, # of fully mapped reads, and read depth. All means using all the five features.