Omics experiments and algorithms

Question 1

Timeseries experiments

Answer 1

• Take a cell or tissue sample
• Apply some change to the environment
• Take n samples at given time points
• Measure each sample
• Analyse how things changed over time

Question 2

Cell type experiments

Answer 2

• Take one or more tissue samples
• Extract similar cells by morphology or fluorescent tagging
• Measure each cell group (proteome, metabolome, transcriptome)
• Analyse how things are different between cells

Question 3

Spatial analysis experiments

Answer 3

• Take a tissue sample
• Either carry out in-situ hybridisation to probes or microdissection followed by sequencing/hybridisation
• Measure each sample
• Categorise where the same came from

Question 4

Applications of dendrograms

Answer 4

• Phylogeny
• Clustering biological entities

Question 5

How do we measure distance?

Answer 5

• Number of substitutions
• Estimate the distance given observed differences and apply a nucleotide/amino acid substitution model
• Euclidean/hamming/cosine distance between feature vectors

Question 6

Distance matrix

Answer 6

• An all-against-all matrix which catalogues all scores and measures how far apart all pairs of entities are. All scores on the diagonal must be zero.
• A distance measure can be used as is
• A similarity measure must be inverted in some way

Question 7

Tree clustering algorithms

Answer 7

• Distance-based (UPGMA)
• Maximum parsimony trees
• Maximum likelihood trees

Question 8

What is UPGMA?

Answer 8

• Unweighted Pair-Group Method with Arithmetic mean
• Unweighted: All pairwise distances contribute equally
• Pair-group: Groups are combined in pairs
• Arithmetic mean: Pairwise distances between groups are means to all group members

Question 9

How does UPGMA work?

Answer 9

1. Form a cluster for each leaf node
2. Find the 2 closest clusters given the average distance between those clusters
3. Merge C1, C2 into a single cluster C
4. Form a node for C, connecting it to C1 and C2. Set the age of C as Davg(C1,C2)/2
5. Eliminate columns for C1 and C2 in D, add a row/column for C and compute the average distances between clusters once again
6. Iterate steps 2-5 until you reach a single cluster containing all clusters

Question 10

UPGMA properties

Answer 10

• Time complexity O(n^2 logn)
• A unique tree
• A rooted tree
• An ultrametric tree (all the leaves are equidistant from the root)

Question 11

Node

Answer 11

A vertex which represents an entity that we wish to model that can have a defined relationship with other nodes

Question 12

Edge

Answer 12

A connection between two nodes that specifies some relationship between them

Question 13

Adjacency

Answer 13

Two nodes are adjacent if connected by an edge

Question 14

Typical experimental design

Answer 14

1. Time-series transcriptomics
2. Data pre-processing
3. Inference methods
4. Network inference
5. Validation
6. Modelling/simulations

Question 15

CLR algorithm

Answer 15

• Take all transcription data
• Calculate mutual information between expression levels of all pairs of genes
• Build MI matrix
• Calculate the z-score for each putative transcription factor and putative target
• Calculate joint z score
• Accept any zi,j that is above a given threshold indicating regulation