Too many "Undetermined Read" on HiSeq

[AndyCai]

HiSeq 2500 Rapid run mode, PE 100bp, on-board cluster generation, dual index 8+8, DNA libraries prepared by Agilent QXT kit.

We had several runs with high percentage undetermined read. Normally it is around 2~3% but in those runs they were 20~30%. The % Perfect Index Reads dropped from ~97% to 60~70%, and the % One Mismatch Reads (Index) increased from 2-3% to 30~40%. When we use the same sequencer to repeat the same sample, the repeat run data was good as normal.

We carefully checked the index sequence of those "undetermined read", and found the index 1 sequences are consistent with some samples in the pool, but index 2 sequences have some mismatched bases.

We don't know if the flowcell has defects or the PE cluster kit has problem. Dose anyone have the similar experience? Thanks a lot!

[GenoMax]

What was the cluster concentration? This sounds like the classic case of a flowcell that is overloaded (but not so much as to lose entire index 2 reads). Were you able to recover data allowing mismatches in the tag read?

[AndyCai]

What was the cluster concentration? This sounds like the classic case of a flowcell that is overloaded (but not so much as to lose entire index 2 reads). Were you able to recover data allowing mismatches in the tag read?

Actually the bad runs had low cluster density:
Lane Tiles Density (K/mm2)
1 64 949 +/- 61
2 64 797 +/- 38
1 64 1008 +/- 45
2 64 1040 +/- 64
1 64 905 +/- 63
2 64 849 +/- 41
1 64 864 +/- 34
2 64 732 +/- 41
1 64 972 +/- 49
2 64 919 +/- 50

The normal runs (good runs) are like this:
1 64 1127 +/- 96
2 64 1129 +/- 96
1 64 1208 +/- 37
2 64 1161 +/- 29
1 64 1254 +/- 92
2 64 1254 +/- 91
1 64 1218 +/- 88
2 64 1237 +/- 89
1 64 1156 +/- 79
2 64 1202 +/- 69

The index read2 %Q30 decreased sharply to 20~60%, but read1, index read1, read2 looked fine (>85%), although slightly lower than normal (>90%).

The cluster PF of bad runs are 84~87%, normal runs are 88~93%.

We didn't try allowing mismatch index. We have to use perfect match reads only.

Thanks!

[jdk787]

Have you contacted Illumina about this?
Like GenoMax stated, low quality index reads are normally a result of overclustering, so I would be interested in hearing what Illumina says about what you're seeing.