1. Marius Nicolae Computer Science and Engineering Department University of Connecticut Joint work with Serghei Mangul, Ion Mandoiu and Alex Zelikovsky

2.  Introduction  EM Algorithm  Results  Conclusions and future work

3. ABCDE Make cDNA & shatter into fragments Sequence fragment ends Map reads Gene Expression (GE) ABC AC DE Isoform Discovery (ID) Isoform Expression (IE)

4.  Read ambiguity (multireads)  What is the gene length? ABCDE

5.  Ignore multireads  [Mortazavi et al. 08] ◦ Fractionally allocate multireads based on unique read estimates  [Pasaniuc et al. 10] ◦ EM algorithm for solving ambiguities  Gene length: sum of lengths of exons that appear in at least one isoform  Underestimate expression levels for genes with 2 or more isoforms [Trapnell et al. 10]

6. ABCDE AC

7.  [Jiang&Wong 09] ◦ Poisson model, single reads only  [Li et al.10] ◦ EM Algorithm, single reads only  [Feng et al. 10] ◦ Convex quadratic program, pairs used only for ID  [Trapnell et al. 10] ◦ Extends Jiang’s model to paired reads ◦ Fragment length distribution

8.  EM Algorithm for IE ◦ Single and paired reads ◦ Fragment length distribution ◦ Strand information ◦ Base quality scores  Solving GE by adding isoform levels

9.  Introduction  EM Algorithm  Results  Conclusions and future work

11.  Paired reads  Single reads ABC AC ABC ACAC ABCABC AC ABC AC ABC AC

12. E-step M-step

13.  Introduction  EM Algorithm  Results  Conclusions and future work

14.  Human genome UCSC known isoforms  GNFAtlas2 gene expression levels ◦ Uniform/geometric expression of gene isoforms  Normally distributed fragment lengths ◦ Mean 250, std. dev. 25

15.  Error Fraction (EF) ◦ Percentage of isoforms (or genes) with relative error larger than given threshold t  Median Percent Error (MPE) ◦ Threshold t for which EF is 50% r2r2 ◦ Coefficient of determination

16.  30M single reads of length 25  Main difference b/w IsoEM and RSEM is fragment length modeling

17.  30M single reads of length 25

18.  Fixed sequencing throughput (750Mb)  50bp reads better than 100bp!

19.  1-60M 75bp reads  Pairs help, strand info doesn’t  [Trapnell et al. 10] r 2 =.95 for 13M PE reads

20.  Introduction  EM Algorithm  Results  Conclusions and future work

21.  Presented EM algorithm for isoform frequency estimation that exploits fragment length distribution for both single and paired reads ◦ Significant accuracy improvement over existing methods ◦ Code and datasets to be released publicly soon  Ongoing extensions ◦ Confidence intervals ◦ Allelic specific isoform expression ◦ Testing for novel isoforms ◦ Integration with isoform discovery