This workshop will address critical issues related to Transcriptomics data: Processing raw Next Generation Sequencing (NGS) data: 1. Next Generation Sequencing data preprocessing: Trimming technical sequences Removing PCR duplicates 2. RNA-seq based quantification of expression levels: Conventional pipelines (looking at known transcripts) Identification of novel isoforms Analysis of Expression Data Using Machine Learning: 3. Unsupervised analysis of expression data: Principal Component Analysis Clustering 4. Supervised analysis: Differential expression analysis Classification, gene signature construction 5. Gene set enrichment analysis The workshop will include hands-on exercises utilizing public domain datasets: breast cancer cell lines transcriptomic profiles (https://genomebiology.biomedcentral.com/articles/10.1186/gb-2013-14-10-r110), patient-derived xenograft (PDX) mouse model of tumor and stroma transcriptomic profiles (http://www.oncotarget.com/index.php?journal=oncotarget&page=article&op=view&path[]=8014&path[]=23533), and processed data from The Cancer Genome Atlas samples (https://cancergenome.nih.gov/). Team: The workshops are designed by the researchers at the Tauber Bioinformatics Research Center at University of Haifa, Israel in collaboration with academic centers across the US. Technical support for the workshops is provided by the Pine Biotech team. https://edu.t-bio.info/a-critical-approach-to-transcriptomic-data-analysis/

2. Special Thank you to:Dr. Vladimir Galatenko, Chief Scientist at the
Tauber Bioinformatics Research Center. His work is
focused on issues related to Big Data analysis and,
in particular, on integration of multi-omics datasets.
A special research interest of Dr. Galatenko is
related to feature selection which is vital for efficient
development of clinical test systems.
Julia Panov, Ph.D. student involved in a number of
neuroscience research projects, an experienced
bioinformatics user. She relies on the T-BioInfo
platform for regular processing and integration of
omics data, collaborating with TBRC research
group on platform development. Dr. Javeed Iqbal, UNMC

3. Biological Examples and Reference Data sets:
• “Modeling precision treatment of breast cancer”, Daemen et. al.
(https://genomebiology.biomedcentral.com/articles/10.1186/gb-2013-14-10-r110),
• “Whole transcriptome profiling of patient-derived xenograft models as a tool to identify both
tumor and stromal specific biomarkers” Bradford et. al.
(http://www.oncotarget.com/index.php?journal=oncotarget&page=article&op=view&path[]=8014
&path[]=23533), and
• Processed data from The Cancer Genome Atlas samples (https://cancergenome.nih.gov/).

4. 1. Next Generation Sequencing data pre-processing:
• Trimming technical sequences
• Removing PCR duplicates
2. RNA-seq based quantification of expression levels:
• Conventional pipelines (looking at known transcripts)
• Identification of novel isoforms
Processing of NGS data:

5. Gene set enrichment analysis

6. Analysis of Expression Data Using Machine Learning:
3. Unsupervised analysis of expression data:
• Principal Component Analysis
• Clustering
4. Supervised analysis:
• Differential expression analysis
• Classification, gene signature construction

7. Part 1:
Biological Significance

9. Cell Line DataTypes: Gene Expression and RNA-editing
RNA-EditingGene expression

10. Breast cancer and cell line models

11. Cell Lines as Cancer Models

12. Sample 1 Sample 2 Sample 3 Sample 4
gene 1 4 3 3 7
gene 2 6 5 5 8
gene 3 6 6 6 6
gene 4 1 2 1 2
gene 5 9 10 1 5
gene 6 12 4 0 5
gene 7 1 7 9 8
gene 8 4 8 3 10
Gene ExpressionTable
Chr Pos start End Sample 1 Sample 2 Sample 3 Sample 4
chr1 1312400 1312400 0 0 0 0
chr1 8362100 8362100 0 0 0 0
chr11 842700 842700 0.705023 0 0 1.17938
chr12 753200 753200 0 0 0 0
chr16 521100 521100 0 0 0 0
chr16 1362700 1362700 0 0 0 0
chr16 1446900 1446900 0 0 8.55549 0
chr16 2176500 2176500 0 0 0 0
chr16 2896600 2896600 0 0 0 0
chr16 29972700 29972700 0 0 0 0
chr16 30358600 30358600 0 0 0 0
chr16 30778800 30778800 0 0 0 0
chr17 2042900 2042900 0 0 15.332 0
chr17 4538300 4538300 0 0 0 0
chr17 4891100 4891100 0 0 0 0
chr17 4946300 4946300 0 38.4794 0 0
chr17 5033200 5033200 0 0 0 0
RNA-editingTable
49 Cell Lines

13. Samples
Genes
Expression
values
Samples
Abundance
values
RNA-editing
Link1
Link2

14. Matrix of distances between samples based on
Gene Expression
Matrix of distances between samples based on
Abundance of RNA editing
HCC202

15. Gene expression and RNA editing abundance
tables similarly separate HCC202 sample

16. Genes and RNA editing
Genes: RNA Editing:
Olfactory
Receptor
s
miRNA
Rab
GTPases
EnhancedTrafficking

17. Learn more at: T-bio.info

18. Part 2:
Working with RNA-Seq Data

19. RNA-seq: overview
.…TCTGAAACAATGCTTCAATCTAACTTATCATTCATTGGGA….
Genome
19

20. RNA-seq: overview
Genome
Gene A Gene B Gene C
20

21. RNA-seq: overview
Genome
Gene A Gene B Gene C
Transcr. ATranscr. ATranscr. A Transcr. ATranscr. C
21

22. RNA-seq: overview
Genome
Gene A Gene B Gene C
Transcr. ATranscr. ATranscr. A Transcr. ATranscr. C
Reads
22

23. RNA-seq: overview
Genome
Gene A Gene B Gene C
Transcr. ATranscr. ATranscr. A Transcr. ATranscr. C
Reads
Transcr. A Transcr. C
23

24. RNA-seq: some details
Genome
Gene A Gene B Gene C
Transcr.Transcr.Transcr. A
Shattering
24
Transcr. CTranscr. C

25. RNA-seq: some details
Genome
Gene A Gene B Gene C
Transcr.Transcr.Transcr. A Transcr.
Adapters ligation
25
Transcr. C

26. RNA-seq: some details
Genome
Gene A Gene B Gene C
Transcr.Transcr.Transcr. A Transcr.
PCR amplification
26
Transcr. C

27. RNA-seq: some details
Genome
Gene A Gene B Gene C
Transcr.Transcr.Transcr. A Transcr.
“Reading”
27
Transcr. C

28. RNA-seq: per-sample processing
Preprocessing:
• Adapters removal plus additional trimming
• Removing PCR duplicates
Mapping
• Mapping on the set of known transcripts
• Mapping on genome (and potential identification of novel transcripts)
• Combined strategy
Quantification of expression levels
28

29. RNA-seq: Comments
PCR removal should be used with caution to avoid removing natural
duplicates (valuable links:
http://www.cureffi.org/2012/12/11/how-pcr-duplicates-arise-in-next-generation-sequencing/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4965708/ - DNA-seq and variant calling
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4597324/ - RNA-seq, ChIP-seq data
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3871669/ - trimming
29

30. RNA-seq: processing
30

31. 31
RNA-seq: processing

32. RNA-seq: expression level quantification
Standard measures
• read counts (raw, expected)
• FPKM – fragments per kilo base per million mapped reads:
Number of reads mapped on the gene /
((total number of mapped reads – in millions) x (gene length – in kilobases))
• TPM – transcripts per million
For one sample TPMg = C x FPKMg, where C is selected in such a way that sum of all
TPMg is one million. But constants C are different for different samples.
32

33. RNA-seq: expression level quantification
Alternative definition of TPM:
(Number of reads mapped on the gene x read mean length x 106) /
(gene length x T),
where T is the sum over all genes of
(Number of reads mapped on the gene x read mean length) /
gene length
Each term here represents the number of sampled transcripts corresponding to a gene, and T estimates the
total number of sampled transcripts (molecules). Thus, TPM is the estimate of the number of transcripts
corresponding to a gene in every million transcripts.
Details: Wagner G.P., Kin K., Lynch V.J. (Theory Biosci., 2012) https://www.ncbi.nlm.nih.gov/pubmed/22872506
33

34. RNA-seq: expression level quantification
Linear scale vs Log-scale
Relative differences are biologically more meaningful than absolute.
Computations are simplified if a log-scaling is performed:
Log-scaled measure = log2 (linear-scale measure + shift)
For relatively large values a difference equal to 1 in log-scale is a 2x difference in linear
scale; difference equal to 3 in log-scale is a 8x difference in linear scale, etc.; difference
equal to -1 in log-scale is a 2x difference in linear scale, but in the opposite direction.
34

35. Comparison: the role of preprocessing
No
preprocessing
35

36. Comparison: the role of preprocessing
No PCR
duplicate
removal
36

37. Comparison: the role of preprocessing
Standard
37

38. Comparison: the role of preprocessing (output)
38

39. Comparison: the role of preprocessing
39

40. Comparison: the role of preprocessing
40

41. Extended pipeline
41

42. Extended pipeline
42

43. BREAK
43
B R E A K

44. Unsupervised analysis: PCA
44

45. Unsupervised analysis: PCA
45

46. Unsupervised analysis: PCA
46

47. 47
Unsupervised analysis: hierarchical clustering

48. 48
Unsupervised analysis: hierarchical clustering

49. 49
Unsupervised analysis: hierarchical clustering

50. Unsupervised analysis: hierarchical clustering
50

51. Unsupervised analysis: hierarchical clustering
51

52. 52
Unsupervised analysis: hierarchical clustering

53. 53
Unsupervised analysis: hierarchical clustering

54. 54
Unsupervised analysis: hierarchical clustering

55. 55
Unsupervised analysis: hierarchical clustering
Dendrogram

56. 56
Unsupervised analysis: hierarchical clustering
Dendrogram

57. Unsupervised analysis: PCA (15 genes)
57

58. Unsupervised analysis: PCA (15 genes)
58

59. Unsupervised analysis: hierarchical clustering, 15 genes
Dendrogram
59

60. Unsupervised analysis: hierarchical clustering, 15 genes
N-like BasalC-lowLuminal 60
Dendrogram

61. Gene annotation: ENSG to Gene Symbols plus GO
61

62. 62
Unsupervised analysis: K-means, 15 genes

63. 63
Unsupervised analysis: K-means, 15 genes

64. 64
Unsupervised analysis: K-means, 15 genes

65. 65
Unsupervised analysis: K-means, 15 genes

66. 66
Unsupervised analysis: K-means, 15 genes

67. 67
Unsupervised analysis: K-means, 15 genes

68. 68
Unsupervised analysis: K-means, 15 genes

69. 69
Unsupervised analysis: K-means, 15 genes

70. 70
Unsupervised analysis: K-means, 15 genes

71. 71
Unsupervised analysis: K-means, 15 genes

72. Unsupervised analysis: K-means, 15 genes
72

73. Unsupervised analysis: K-means, 15 genes
“The SUM52PE cell line was derived from a pleural effusion and was found to be
negative for ER and PR expression, however the original primary tumor from this
patient was positive for both hormone receptors”.
Chavez KJ, Garimella SV, Lipkowitz S. Triple negative breast cancer cell lines: one tool in the
search for better treatment of triple negative breast cancer. Breast Dis. 2010; 32(1-2):35-48.
Ethier SP, Kokeny KE, Ridings JW, Dilts CA. erbB family receptor expression and growth regulation
in a newly isolated human breast cancer cell line. Cancer Res. 1996; 56(4): 899-907.
73

74. BREAK
74
B R E A K

75. 75
Supervised analysis:
SVM with a linear kernel as an example

76. 76
Supervised analysis:
SVM with a linear kernel as an example

77. 77
Supervised analysis:
SVM with a linear kernel as an example

78. d
d
78
Supervised analysis:
SVM with a linear kernel as an example

79. 79
Supervised analysis:
SVM with a linear kernel as an example

80. ?
80
Supervised analysis:
SVM with a linear kernel as an example

81. Supervised analysis:
SVM with a linear kernel as an example
?
81

82. Supervised analysis: available methods
• Linear Discriminant Analysis (LDA)
• Quadratic Discriminant Analysis (QDA)
• Random Forest
• Support Vector Machine (SVM)
• Naïve Bayes
82

83. Supervised analysis: 15 genes
83

84. Differential expression analysis
Quantities related to the degree of differential
expression:
• Difference between mean expression levels – fold
change (please, pay attention to scale);
• Statistical significance – p-value, adjusted p-value
(e.g., FDR)
• Expression level magnitude (caution with low-
expressed genes from the analysis).
84

85. Differential expression analysis
85

86. Differential expression analysis
86

87. Gene set / pathway enrichment analysis
Possible options:
• Use only lists (thresholding required): one of the standard
tools here is The Database for Annotation, Visualization and
Integrated Discovery – DAVID
(https://david.ncifcrf.gov/home.jsp, https://david-
d.ncifcrf.gov/).
• Take into consideration degrees of differential expression;
• Additionally take into consideration pathway topology.
87

88. Gene set / pathway enrichment analysis
88

89. Gene set / pathway enrichment analysis
89

90. BREAK
90
B R E A K

91. BREAK
91
HANDSON
Separation of TCGA and breast
cancer PDX samples

92. BREAK
92
HANDSON
Analysis of a subset of breast
cancer PDX samples