Table 3.

Performance of RNAIndel on two pediatric and three adult cancer datasets

	Tumor	N	Library	ReadLen	Sequencer	SomaticIndels*		TPR	Median #FP/Sample^			Median indels per sample
									A	B	C
1	Pediatric	77	TotalRNA	100	HiSeq2000or 2500	s	17	0.882	3		3	2318
						m	40	0.975	4		3	311
2	AML	158	Poly-A	75	HiSeq2000	s	22	0.954	2		1	1036
						m	61	0.984	2		2	202
3	NSCLC	90	Poly-A	100	HiSeq1500	s	97	0.887	6	3	4	3171
						m	68	0.941	7	4	4	394
4	RCC	91	Poly-A	50	HiSeq2000	s	130	0.877	8	5	5	4303
						m	81	0.889	8	2	2	510
5	COAD (Hyper)	29	Poly-A	75	GAIIX	s	120	0.392	20		20	999
						m	53	0.953	11		10	141
	COAD (NonHyper)	102				s	30	0.9	4		3	466
						m	14	1.000	4		2	128

Note: TestSet 1 consists of 77 samples from 20 types of pediatric cancers (Supplementary Table S2). The tumor type for TestSet 2–5 uses the following abbreviation: AML for acute myeloid leukemia; NSCLC for non-small cell lung cancer; RCC for renal cell carcinoma; COAD for colon adenocarcinoma with hypermutator phenotype (Hyper) or without hypermutator phenotype (NonHyper). *s for single-nucleotide indel, m for multi-nucleotide indel. ^{^}Median number of false positives (FP) in somatic prediction per sample is shown for the default (column A), filtered with normal RNA-Seq data (column B, available only for TestSet 3 and 4) and filtered with cohort recurrence in RNA-Seq (Supplementary Fig. S4).