2. T-BioInfo is designed for processing, analysis and
integration of multi-omics data. The platform is used in
multiple research groups to extract meaningful insights
from large multi-omics datasets. Our current effort
expands to education, by enabling more people to
extract meaningful, data-driven insights from omics
datasets with biomedical applications. To learn more
about the platform and it’s research and educational
features, follow the highlighted links .
T-bio.info | edu.t-bio.info | server.t-bio.info
2
9. Daemen et al., 2013, “Modeling precision treatment of breast cancer”: an analysis of over 70 different Breast Cancer cell lines and over 90 different
therapeutic agents. https://genomebiology.biomedcentral.com/articles/10.1186/gb-2013-14-10-r110 9
16. Preprocessing:
• Adapters removal plus additional
• Removing PCR duplicates
16
Quantification of expression levels
Mapping
• Mapping on the set of known transcripts
• Mapping on genome (and potential
identification of novel transcripts)
• Combined strategy
RNA-Seq: overview
21. Standard Measures of RNA Quantification:
• Counts
• FPKM – fragments per kilobase 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
million. Constants C are different for different samples.
21
22. Linear scale vs Log-scale
Relative differences are biologically more meaningful than absolute.
are simplified if a log-scaling is performed:
Log-scaled measure =
log2 (linear-scale measure + shift)
For relatively large values:
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.
22
23. Preprocessing:
• Adapters removal plus additional
• Removing PCR duplicates
23
Quantification of expression levels
Mapping
• Mapping on the set of known transcripts
• Mapping on genome (and potential
identification of novel transcripts)
• Combined strategy
RNA-Seq: overview
24. Comparison: the role of preprocessing
24
High expression can be affected by pre-processing steps like PCR-clean and “Trimmomatic”
33. 33
Unsupervised analysis: Hierarchical Clustering
• Identify groups
• Associate sample to group
Why use clustering?
• Various methods
• Random selection in some methods
• Interpretation
Considerations:
43. • Fitting classifier on training set and predicting classes on the test set
• Is it possible to tune 7000 coefficients by 52 samples?
• Some algorithms do feature selection: swLDA, random forest
• Other algorithms won’t work if number of features >> number of
samples
• Curse of dimensionality
43
Considerations Supervised analysis
44. 44
• Extracting 15 highly informative genes from the swLDA classifier
• How other supervised learning algorithms can be applied (e.g.,
SVM)
• Feature selection can also improve quality of unsupervised learning
analysis
Step-wise Linear Discriminant Analysis (swLDA)
45. 45
Classification Practice
• Organize the table with 15
genes by sample type
• Color expression (green –
low; red – high)
• Which genes stand out?
• Which sample stand out?
• What groups are hard to
detect?
CellLines_15Genes_market.txt
56. 56
Part 1: Conventional Machine Learning Approaches for Next
Generation Sequencing
Rapid RNA-seq processing for expression quantification applying
logical pipeline construction and pre-processing considerations.
hands-on exercises, participants will explore the expression
using conventional unsupervised machine learning methods and
supervised classifiers with and without feature extraction. Using
BioInfo platform, participants will learn about the logic and
considerations of applying such methods and be prepared for
independent downstream analysis and visualization of data
downloaded R scripts produced by the system. The
produced/downloaded code will be reviewed, customized and
subsequent session.
T-bio.info | edu.t-bio.info (FREE) | server.t-bio.info (14 days DEMO)
58. 58
Required installations:
R >= 3.4
R Studio
gplots
ggfortify
ggplot2
ggpubr
e1071
mda
MASS
klaR
Part 2: Combining custom software with R to
streamline analysis workflows and visualize ‘Omics
data insights.
Differential Gene Expression, Gene Set Enrichment
Analysis
R visualization from scratch: utilize the same dataset for
basic data exploration and visualization in R.
This session will strengthen the participants ability to
transition to script-based workflows in RNA-seq
downstream analysis and visualization. Participants will
learn about downstream capabilities of R-based workflow
to transform and manipulate tables and visualize findings
in a meaningful way.
61. 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)
• Level of Expression (caution with low-expressed genes from the
analysis)
61
62. • Hard to interpret when number of groups is greater than two, so we can use Claudin-low vs normal-
like groups.
• Differential Expression is a natural and easy to interpret feature selection procedure.
• Pathway enrichment analysis can be applied to the resulting table 62
Differential expression analysis
65. Gene set / pathway enrichment analysis
GAGE -
• Use only lists (thresholding required): one of the standard tools here isThe
Database for Annotation,Visualization and Integrated Discovery – DAVID
(https://david.ncifcrf.gov/home.jsp, https://david-d.ncifcrf.gov/).
• Takes into consideration level of differential expression
65
68. 68
Gene set / pathway enrichment analysis
Regulation of Actin Cytoskeleton B Cell Receptor Signaling Pathway
69. 69
Required installations:
R >= 3.4
R Studio
gplots
ggfortify
ggplot2
ggpubr
e1071
mda
MASS
klaR
Part 2: Combining custom software with R to
streamline analysis workflows and visualize ‘Omics
data insights.
Differential Gene Expression, Gene Set Enrichment
Analysis
R visualization from scratch: utilize the same dataset for
basic data exploration and visualization in R.
This session will strengthen the participants ability to
transition to script-based workflows in RNA-seq
downstream analysis and visualization. Participants will
learn about downstream capabilities of R-based workflow
to transform and manipulate tables and visualize findings
in a meaningful way.