Incorrect codon files generated by scikit-ribo-build for mm10

[nabeel-bioinfo]

I have been trying to use scikit-ribo on mouse ESC Ribo-Seq data but the output files generated from scikit-ribo-build do no correctly parse the given input GFF file. I am using the mm10_knownGene GFF file as input with genome file obtained from http://hgdownload.cse.ucsc.edu/goldenPath/mm10/chromosomes/. Specifically, there is inaccurate assignment of codons of transcripts.

For example, for gene uc008cjg.1, the start codon is at chr17: 35916361-35916363 while in the files mm10.codons.df and mm10.codons.bed, this chromosome location is at codon 2 $ cat mm10.codons.bed | grep 'uc008cjg.1' | head -12

chr17	35916390	35916393	uc008cjg.1	-8	-	CGG
chr17	35916387	35916390	uc008cjg.1	-7	-	CAG
chr17	35916384	35916387	uc008cjg.1	-6	-	GCC
chr17	35916381	35916384	uc008cjg.1	-5	-	GTA
chr17	35916378	35916381	uc008cjg.1	-4	-	CGT
chr17	35916375	35916378	uc008cjg.1	-3	-	CCG
chr17	35916372	35916375	uc008cjg.1	-2	-	GTG
chr17	35916369	35916372	uc008cjg.1	-1	-	TGC
chr17	35916366	35916369	uc008cjg.1	0	-	GTC
chr17	35916363	35916366	uc008cjg.1	1	-	AAG
chr17	35916360	35916363	uc008cjg.1	2	-	ATG
chr17	35916357	35916360	uc008cjg.1	3	-	GCT

Another example where the codon assignments are entirely different from the reference is uc007vnb.1. The start codon is at chr15:37003817-37003820 but the codon file has very different annotations.

$cat mm10.codons.bed | grep 'uc007vnb.1' | head -20

chr15	37007423	37007426	uc007vnb.1	-8	-	CCG
chr15	37007420	37007423	uc007vnb.1	-7	-	CAG
chr15	37007417	37007420	uc007vnb.1	-6	-	GAA
chr15	37007414	37007417	uc007vnb.1	-5	-	GGT
chr15	37007411	37007414	uc007vnb.1	-4	-	GAG
chr15	37007408	37007411	uc007vnb.1	-3	-	GGG
chr15	37007405	37007408	uc007vnb.1	-2	-	CGG
chr15	37007402	37007405	uc007vnb.1	-1	-	GAC
chr15	37007399	37007402	uc007vnb.1	0	-	CGC
chr15	37007396	37007399	uc007vnb.1	1	-	GCG
chr15	37007393	37007396	uc007vnb.1	2	-	GGC
chr15	37007390	37007393	uc007vnb.1	3	-	AAG

Assuming that codon 0 is the start codon in this file, I found that only 3380(7%) of 46355 transcripts have ATG as start codon

cat mm10.codons.bed | awk '{if($5==0)print $5,$7}' | sort | uniq -c

727 0 AAA
    436 0 AAC
    748 0 AAG
    376 0 AAT
    686 0 ACA
    473 0 ACC
    295 0 ACG
    854 0 ACT
   1768 0 AGA
   1212 0 AGC
   1207 0 AGG
   1724 0 AGT
    377 0 ATA
    522 0 ATC
   **3380 0 ATG**
    704 0 ATT
    228 0 CAA
    377 0 CAC
    493 0 CAG
    167 0 CAT
    400 0 CCA
    569 0 CCC
    399 0 CCG
    477 0 CCT
    188 0 CGA
    602 0 CGC
    637 0 CGG
    136 0 CGT
    202 0 CTA
    831 0 CTC
    625 0 CTG
    594 0 CTT
    990 0 GAA
    779 0 GAC
   1888 0 GAG
    585 0 GAT
    921 0 GCA
   1177 0 GCC
   1318 0 GCG
    928 0 GCT
   1856 0 GGA
   1691 0 GGC
   2329 0 GGG
   1037 0 GGT
    523 0 GTA
   1015 0 GTC
   1304 0 GTG
    916 0 GTT
     90 0 TAA
     88 0 TAC
    184 0 TAG
    115 0 TAT
    284 0 TCA
    401 0 TCC
    162 0 TCG
    386 0 TCT
    285 0 TGA
    392 0 TGC
    460 0 TGG
    270 0 TGT
    155 0 TTA
    506 0 TTC
    274 0 TTG
    632 0 TTT

I hope this bug can be fixed so that scikit-ribo can be run on mm10 data.