1. High-throughput omic datasets and clustering
Colin Dewey
BMI/CS 576
Fall 2015

2. Goals for today Recap of molecules of life
High-throughput datasets/omic datasets
Clustering approaches to analyzing high-throughput datasets
What is clustering?
Applications to mRNA/gene expression levels

3. Molecules of life DNA RNA Proteins Metabolites
mRNA
ncRNA
Proteins
Metabolites
While DNA is mostly static, epigenome, RNA, proteins, metabolites change between cell types, tissues, environments and conditions
The central dogma

4. High-throughput datasets and “omes”
Aim to measure as many components of a cells simultaneously
Types of omes
Genome: collection of DNA in a cell
Epigenome: all of the chemical modifications on the genome
Transcriptome: all of the RNA in cell
Proteome: all of the proteins in a cell
Metabolome: all of the metabolites present in a cell
Interactome: all of the interactions within a cell

5. Omics data provide comprehensive description of nearly all components of the cell
Joyce & Palsson, Nature Mol cell biol. 2006

6. Databases with omic data
Joyce & Palsson, Nature Mol cell biol. 2006

7. Understand a cell as a system
Measure: identify the parts of a system
Parts: different types of bio-molecules
genes, proteins, metabolites
High-throughput assays to measure these molecules
Model: how these parts are put together
Clustering
Network inference and analysis

8. Transcriptomes/genome-wide mRNA levels are dynamic
mRNAs
genes
What is varied: individuals, strains, cell types, environmental conditions, disease states, etc.
What is measured: RNA quantities for thousands of genes, exons or other transcribed sequences

9. Bio-techniques to measure transcriptomes
Microarrays
cDNA/spotted arrays
Oligonucleotides arrays
Sequencing
RNA-seq

10. Microarrays
A microarray is a solid support, on which pieces of DNA are arranged in a grid-like array
Each piece is called a probe
Measures RNA abundances by exploiting complementary hybridization
DNA from labeled sample is called target

11. Microarrays

12. Complementary hybridization
Due to Watson-Crick base pairing, complementary single-stranded DNA/RNA molecules hybridize (bond to each other)
AGCGGTTCGAATACC
TCGCGAAGCTAGACA
CCGAAATAGCCAGTA
UCGCCAAGCUUAUGG

13. Complementary hybridization
one way to do it in practice
put (a large part of ) the actual gene DNA sequence on array
convert mRNA to cDNA using reverse transcriptase
TCGCCAAGCTTATGG
actual gene
AGCGGTTCGAATACC
cDNA
reverse transcriptase
UCGCCAAGCUUAUGG
mRNA

14. Spotted vs. oligonucleotide arrays
spotted arrays:
synthesize samples of cDNA (full-length transcripts or shorter sequences) and then spot them onto array
Usually two colors
oligonucleotide arrays:
synthesize sets of DNA oligonucleotides (short, fixed length sequences, typically nucleotides in length) on array itself (in situ)
Commercially available
Affymetrix
Nimblegen
Usually one color
In both cases, mRNA is converted to DNA, labeled and hybridized, and detected by fluorescence scanning

15. Spotted versus oligonucleotide array
From Vermeeren and Luc Michiels 2011 DOI: /19432

16. A video about two color DNA microarrays

17. Microarray measurements
Can’t detect the absolute amount of mRNA present for a given gene, but can measure a relative quantity
For two color arrays, the measurements represent
where red is the test expression level, and green is the reference level for gene G in the i th experiment

18. RNA-seq measurements
Measurements are digital: counts of sequenced reads for each gene/transcript
Still the measurements represent relative amounts of each transcript: the counts depend on how many reads are sequenced

19. A typical RNA-seq pipeline
Wang et al, Nature Genetics 2009

20. RNA-seq vs. microarrays
Advantages of RNA-seq
Don’t need reference sequence for genes/genome being assayed
Low background noise
Large dynamic range (105 vs. 102 for microarrays)
High technical reproducibility
Disadvantage
more expensive, but cost is rapidly falling

21. Gene expression profiles
We will assume we have a 2D matrix of gene expression measurements
rows represent genes
columns represent different experiments, time points, individuals etc.
We will refer to individual rows or columns as profiles
a row is a profile for a gene
a column is a profile for an experiment, time point, etc.

22. Gene-expression profiles for yeast cell cycle
Rows represent yeast genes
Columns represent time points as yeast goes through cell cycle
Color represents expression level relative to baseline (red=high, green=low, black=baseline)
Spellman 1998

23. Gene-expression profiles for leukemia patients
rows represent genes
columns represent people with 2 subtypes of leukemia: ALL and AML
Each column corresponds to a microarray measurement

24. Gene-expression profiles for genes that induce differentiation
Ivanova et al., Nature 2006

25. Commonly asked questions from expression datasets
If we measure gene expression in a normal versus disease cell type, which genes have different expression levels across two groups?
Differential expression
Which genes seem to be changing together?
Clustering genes based on expression profiles of genes across all conditions
Which treatments/individuals have similar profiles?
Clustering samples based on gene expression profiles of all genes
What does a gene do?
What functional class does a given gene belong
What class is a sample from?
e.g., does this patient have ALL or AML
How will this sample react to a particular drug?

26. Clustering of gene expression data
Task definition
Distance metric
Flat clustering
K-means
Model-based clustering
Gaussian mixture models
Hierarchical clustering
Top-down and bottom-up

27. Task definition: clustering gene expression profiles
Given: expression profiles for a set of genes or experiments/individuals/time points (whatever columns represent)
Do: organize profiles into clusters such that
profiles in the same cluster are highly similar to each other
profiles from different clusters have low similarity to each other

28. Motivation for clustering
Exploratory data analysis
understanding general characteristics of data
visualizing data
Generalization
infer something about an object (e.g. a gene) based on how it relates to other objects in the cluster

29. Example output of clustering
Environment conditons
Genes
Genes
Clusters
Genes
Gasch & Eisen, 2002

30. The clustering landscape
There are many different clustering algorithms
They differ with respect to several properties
hierarchical vs. flat
hard (no uncertainty about which profiles belong to a cluster) vs. soft clusters
overlapping (a profile can belong to multiple clusters) vs. non-overlapping
deterministic (same clusters produced every time for a given data set) vs. stochastic
distance (similarity) measure used

31. Distance measures
Central to all clustering algorithms is a measure of distance between objects being clustered
Clustering algorithms aim to group “similar” things together
Optimize some measure of within cluster similarity
Defining the right similarity or distance is an important factor in getting good clusters
Most algorithms will work with symmetric dissimilarities
Dissimilarities may not be distances

32. Notation for microarray data
p microarrays/samples
Data matrix:
……… .
Is a tall matrix of numbers with rows corresponding to genes and columns corresponding to conditions
n Genes xi
Expression profile of ith gene
xij
Expression value of ith gene in jth microarray
jth microarray

33. Different dissimilarity measures
Euclidean distance between two vectors xi and xk
Manhattan distance

34. Distance Metrics Properties of a metric or distance function
(non-negativity)
(identity)
(symmetry)
(triangle inequality)

35. Summary
High-throughput biological measurement technology is enabling comprehensive molecular profiling of cells
Often faced with large matrices of such profiling data
Rows = biomolecular entities (e.g., genes)
Columns = samples (e.g., patients or timepoints)
The clustering task is a first step towards understanding such data
Core to clustering is the definition of “similarity” or “distance”