mapping2015

Transcript

1. Does mapping simulated RNA-seq reads provide information? Leonardo Collado-Torres tweet: @fellgernon blog: tinyurl.com/FellBit
2. Previously • Choose 10 genes with FPKM > 20 • cufflinks: estimate isoform FPKM from 7 Geuvadis samples • polyester: simulate with uniform & rnaf models • Map with TopHat • View coverage https://github.com/alyssafrazee/polyester_code/blob/master/polyester_manuscript.Rmd
3. https://github.com/alyssafrazee/polyester_code/blob/master/polyester_manuscript.Rmd
4. Goals • Reproduce • Similar behavior in other genes/samples? • Do simulated reads predict observed data? Get a measure by gene • Even more than by chance?
6. Original
7. Reproduced
12. Measures by gene: using single bp exon • Correlation (R^2) • RMSD: scaling by max first • ARIMA: forecast models – Auto ARIMA on obs – Fit again using simulated data as predictor – P-value for predictor and estimated coef • Chance: – Neg. binomial size 1 and 6 – Compare replicates of simulated data
13. • d: default params, 2 reps (2x) • r: using rnaf bias, 2x • b1: neg. binomial with size = 1, 2x • b6: neg. binom. size = 6, 2x • d1-d2: using d1 as “obs” – Same for r1-r2, b1a-b1b, b6a-b6b
14. Correlation
15. Correlation: summarize by sample
16. Correlation: summarize by gene
17. R^2
18. Scaling by max then RMSD
19. Scaling by max then RMSD
22. ARIMA: predictor p-value
23. ARIMA: predictor p-value
24. ARIMA: predictor coefficient
27. Code • https://github.com/alyssafrazee/polyest er_code/blob/master/polyester_manuscri pt.Rmd • https://github.com/lcolladotor/mapBias (private for now)