I am a little surprised there is not more posting about the performance of your instruments with v3 chemistry. I am pretty happy with results we are seeing on our HiScanSQ (which uses HiSeq chemistry/flow cells). But we do have one irritating thing (unrelated to v3) happening. We are seeing:
(I cropped out lanes 7 and 8 because they did not show the phenomenon.) See how the cluster densities are quite high, >1000K/mm2 for tiles on the lower left of the flow cell. But the cluster densities are quite low for tiles on the upper right, <500K/mm2. Illumina tells you to shoot for 750-850 K/mm2. But with a cluster gradient >2x to deal with that is pretty difficult. (BTW this was caused by the heat block in the cBot leaking coolant!) Would have been nice to know what happens >850 K/mm2.
Anyway, let me put all whinging aside because our little HiScanSQ appears to be generating sequence at a rate that exceeds that of the entire world circa 5 years ago. Instead I thought I would use this as an opportunity to assess the effect of cluster density on cluster PF and %Q30. Obviously "your mileage may vary", this is N=1 run data from a single instrument. (Also note that higher than optimal density has greater effect after turn around -- see reply to this post for more details.)
The first cut to your data is the one where the software doesn't see the raw cluster at all because it is packed in too tight with other clusters that have the same first 4 bases of sequence. We don't have those sorts of libraries in this run and have kept the adapter-dimer levels fairly low, so I don't think that plays much of a role.
The next cut is where a raw cluster does not Pass Filter (PF). At very high densities you end up with less PF clusters than at lower densities. So how bad does it look? Here it is:
(The colors denote "lane" in this picture.)
Looks like Illumina played it safe with the <850 K/mm2 figure. Up to 900 K/mm2 there does not yet appear to be a down side. Above 900 K/mm2 strange things happen. First the results are mixed. For some tiles the number of clusters drops substantially. For other tiles, there is a modest increase in PF clusters.
Alright, for the clusters that PF, how does the %Q30 fair?
Color indicates cycle number as denoted in legend. I think the category "5-12" cycles is really "1-12" cycles.
The answer is that if you manage to pass filter, there is only a modest effect on sequence quality -- except for the first 5 cycles. Okay, with that error in the legend by SAV, that might not be clear. If I deliberately exclude the first 5 cycles:
Here, I think, it is obvious that the major impact to %Q30 on PF clusters is in the earliest cycles. Here is another visualization of the impact:
Here, color denotes density. Sorry about the crazy cutoffs!
Again, very high densities impact the %Q30 most in the early cycles and, to a lesser extent, in the later cycles.
Why? My guess would be that in the early cycles cluster signal bleed over is highest. Note that the cycle 51 bump in intensities may be the cause of that cycle 51 drop in %Q30.
--
Phillip
(I cropped out lanes 7 and 8 because they did not show the phenomenon.) See how the cluster densities are quite high, >1000K/mm2 for tiles on the lower left of the flow cell. But the cluster densities are quite low for tiles on the upper right, <500K/mm2. Illumina tells you to shoot for 750-850 K/mm2. But with a cluster gradient >2x to deal with that is pretty difficult. (BTW this was caused by the heat block in the cBot leaking coolant!) Would have been nice to know what happens >850 K/mm2.
Anyway, let me put all whinging aside because our little HiScanSQ appears to be generating sequence at a rate that exceeds that of the entire world circa 5 years ago. Instead I thought I would use this as an opportunity to assess the effect of cluster density on cluster PF and %Q30. Obviously "your mileage may vary", this is N=1 run data from a single instrument. (Also note that higher than optimal density has greater effect after turn around -- see reply to this post for more details.)
The first cut to your data is the one where the software doesn't see the raw cluster at all because it is packed in too tight with other clusters that have the same first 4 bases of sequence. We don't have those sorts of libraries in this run and have kept the adapter-dimer levels fairly low, so I don't think that plays much of a role.
The next cut is where a raw cluster does not Pass Filter (PF). At very high densities you end up with less PF clusters than at lower densities. So how bad does it look? Here it is:
(The colors denote "lane" in this picture.)
Looks like Illumina played it safe with the <850 K/mm2 figure. Up to 900 K/mm2 there does not yet appear to be a down side. Above 900 K/mm2 strange things happen. First the results are mixed. For some tiles the number of clusters drops substantially. For other tiles, there is a modest increase in PF clusters.
Alright, for the clusters that PF, how does the %Q30 fair?
Color indicates cycle number as denoted in legend. I think the category "5-12" cycles is really "1-12" cycles.
The answer is that if you manage to pass filter, there is only a modest effect on sequence quality -- except for the first 5 cycles. Okay, with that error in the legend by SAV, that might not be clear. If I deliberately exclude the first 5 cycles:
Here, I think, it is obvious that the major impact to %Q30 on PF clusters is in the earliest cycles. Here is another visualization of the impact:
Here, color denotes density. Sorry about the crazy cutoffs!
Again, very high densities impact the %Q30 most in the early cycles and, to a lesser extent, in the later cycles.
Why? My guess would be that in the early cycles cluster signal bleed over is highest. Note that the cycle 51 bump in intensities may be the cause of that cycle 51 drop in %Q30.
--
Phillip
Comment