Inconsistent library type

[ckr123tw]

Hi, I tried to let Salmon infer library type automatically on TCGA data, however, it seemed to find a high percentage of reads (nearly 40%) inconsistent with the inferred library type. Did I made a mistake in the processing steps? Are the results still valid? Thanks.

Command line: samtools bam2fq 59cd4694-4001-483b-9a30-33a166943bd1_gdc_realn_rehead.bam > 59cd4694-4001-483b-9a30-33a166943bd1.fastq cat 59cd4694-4001-483b-9a30-33a166943bd1.fastq|grep '^@./1$' -A 3 --no-group-separator > 59cd4694-4001-483b-9a30-33a166943bd1_R1.fastq cat 59cd4694-4001-483b-9a30-33a166943bd1.fastq|grep '^@./2$' -A 3 --no-group-separator > 59cd4694-4001-483b-9a30-33a166943bd1_R2.fastq salmon quant -i ~/hg38_ref/GCA_000001405.15_GRCh38_no_alt -l A -1 59cd4694-4001-483b-9a30-33a166943bd1_R1.fastq.gz -2 59cd4694-4001-483b-9a30-33a166943bd1_R2.fastq.gz -p 6 -o ./SalmonQuant

lib_format_counts.json: "read_files": "( 59cd4694-4001-483b-9a30-33a166943bd1_R1.fastq.gz, 59cd4694-4001-483b-9a30-33a166943bd1_R2.fastq.gz )", "expected_format": "MU", "compatible_fragment_ratio": 0.6136629002400855, "num_compatible_fragments": 27796954, "num_assigned_fragments": 45296781, "num_consistent_mappings": 115802038, "num_inconsistent_mappings": 95238168, "strand_mapping_bias": 0.5013267642146333, "MSF": 0, "OSF": 544092, "ISF": 39968962, "MSR": 0, "OSR": 556020, "ISR": 39905960, "SF": 7149723, "SR": 7113411, "MU": 0, "OU": 0, "IU": 0, "U": 0

[bounlu]

I usually get IU for auto-detected library type for TCGA samples.

[rob-p]

Yup, and the fact that this ended up as MU is strange, since the library type frequencies clearly suggest IU (since ISF and ISR counts seem to dominate). Could it be the result of having the FASTQ files generated by converting from BAM which some sort of bias in the beginning reads? The automatic detection uses the first 10,000 reads to decide --- if these are mapped in a biased way, that could be the cause.