Kun:LabNotes/SingleCellExpr/2013-8-14

From ZhangLabWiki
Jump to navigation Jump to search

HL157 run data analysis[edit]

  • Rui&Blue make eight libraries from single mNPCs using various protocols.Rui:LabNotes/SingleCell/2013-7-31; Blue:RNA-Seq_Experiments:07252013
  • The libraries were sequenced on a PE flowcell.
  • There were no reads in Lane 6-8 (three IVT libraries made by Rui).
  • The other IVT library made by Rui has 33 million reads in Lane 5. However, most reads came from the N6 primer annealing on the T20 primer, there was no insert.
  • The following analysis focused on Lane 1-4. Because the read numbers were not high, I did not split each data file into single cells. Instead I treated the pooled libraries as individual data sets to get a quick idea on the quality of the libraries.
  • I used the standard data processing procedures that involve two round of read trimming, tophat mapping, UMI counting and cufflink quantification.
Data set # raw reads # UMI/GGG reads Template switching accuracy # uniquely mapped reads (MQ>20) # unique/non-clonal reads # ERCC reads # unique ERCC reads
s_1_1_Indx14.txt 29,228,570 26,331,150 90.1% 1,873,037 952,715 4,184 822
s_2_1_Indx15.txt 25,678,289 18,851,465 73.4% 1,311,840 718,991 1,727 569
s_3_1_Indx81.txt 12,930,738 5,474,853 42.3% 1,096,219 223,910 214 87
s_4_1_Indx82.txt 13,748,578 6,730,117 49.0% 1,924,080 354,384 114 73
  • A number of observations:
    • Rui's libraries (Indx14/15) have more reads that contain the expected TSO structure. However, there were also much more reads with non-informative poly-A sequences.
    • Blue's libraries have much less poly-A sequences, with the downside that less than half of the reads contain the expected TSO structure, suggesting that something could be wrong with the template switching reactions.
    • All four libraries have 1-2 million uniquely mapped reads. However, there were much fewer reads in Blue's libraries after removing clonal reads based on UMIs. Note that there were probably(??) twice as many input cells in Blue's libraries compared with Rui's. That's an indication that somehow Rui's libraries have higher complexity. The reason could be due to some subtle differences in the protocols, or perhaps different number of PCR cycles used.
    • Somehow there were a lot more reads coming from the ERCC spike-ins in Rui's libraries. Why??? Assuming the efficiency of turning a pure ERCC transcript is constant, then perhaps Blue's protocol led to more complete cell lysis? Or Blue simply started with more cells per aliquot of ERCC spike-in?
    • Can we estimate the number of input RNA molecules/fragments that were converted into the sequencing libraries based on the relative fraction of reads assigned to ERCC versus the mouse genome??