pipeline 1.4 versus 1.3

[dvh]

Thought this might be of interest.

Illumina say the latest version of the Pipeline (v1.4) has better image analysis (Firecrest). To test this claim, I analysed both ways a 70bpx2 PhiX lane from a recent standard GAII flowcell (cycle seq kit v3).

Results indeed suggest lower error rate for Pipeline 1.4 at the same time as generating 20% more data from this cluster density.

I havent yet looked to see if there is any difference in sequencing error types.

Detail below.

David




PIPELINE 1.3
153701 raw clusters/tile
108817 post filter clusters/tile

ELAND 93.39% align, error rate 1.80%

maq-0.7.1 (map -n 3)
[match_data2mapping] 21279032 out of 21755852 raw reads are mapped with
20217141 in pairs.
-- (total, isPE, mapped, paired) = (21755852, 1, 21279032, 20217141)
Error rate 0.025004
Uniquely mapping reads (MQ>30) 20375224


PIPELINE 1.4:
153682 raw clusters/tile
131772 post filter clusters/tile
ELAND: 98.14% align, error rate 1.65%

maq-0.7.1 (map -n 3)
[match_data2mapping] 25751286 out of 26342226 raw reads are mapped with
24449419 in pairs.
-- (total, isPE, mapped, paired) = (26342226, 1, 25751286, 24449419)
Error rate 0.021450
Uniquely mapping reads (MQ>30) 24698897

[bioinfosm]

thanks!

an unrelated question though, what steps do you take to keep the raw clusters/tile constant across runs?

[Bruce E]

We are starting a comparison too. I will post our results. I was wondering if anyone has tried SCS 2.4 yet and if they would care to share their experiences.
Thanks

[basickler]

One important thing to note about 1.4 is that you can also run it at much higher cluster densities. PL 1.3 maxed out at around 185,000-195,000 clusters per tile and PL 1.4 can do 210,000+ clusters per tile and higher.

B

[dwmohr]

We've started a comparison as well, particularly on samples that were clustered at high densities (180,000+). We've gotten between 30 and 50% more data for these samples.

Our initial results suggest that while you may be able to cluster at higher density, your effective yield doesn't increase over samples that are clustered at previously 'ideal' densities.

[basickler]

Ya the usable density is pretty dependent on focus quality and a few other metrics but this is what we've setteled on (using data collected from about a dozen runs on each)

PL. 1.3.2 - 185,000 clusters/tile
RTA/1.3.4 - 210,000 clusters/tile

B

[bioinfosm]

Interestingly we see PF clusters from v1.3 that *fail* PF in v1.4

The reads look alright, and I am not sure why they fail in the new pipeline!

[zlu]

We have recently upgraded our pipeline to from 1.3.2 to 1.4 and the improvement we see when, analyzing a 2x76 cycles run, is in line with what have been said here, i.e. 65% improvement on reads passing chastity filter and nearly double the alignment precentage. However, in 4 of the lanes we have seen a marked increase in the error rate from 0.8 to about 2.3% while the error rates for the other lanes has decreased from about 2.4 to 1.9%. We reckon this might be due to poor libraries prep. since our PhiX control also has an error rate of about 1%. I'll appreciate if anyone can share their views on this. Thank you.