Week 12

Question 1

Clustering method: partitioning

Answer 1

Arbitrarily choose k objects, reassign based on mean of clusters and update until no change.

Question 2

Problems of partitioning clustering

Answer 2

Sensitive to outliers

Takes time

Question 3

Problems of hierarchical clustering

Answer 3

Join unrelated objects
Rigid
Hard to define clusters

Question 4

Data structure based internal measures for number of clusters

Answer 4

maximise inter-distance

minimise intra-distance

Question 5

Stability-based metrics for number of clusters

Answer 5

remove part of information and regenerate until there is no change in clustering results

Question 6

Integrating known biological information for number of clusters

Answer 6

Use a knowledge base

Question 7

Meta-analysis

Answer 7

combine results of published data, check for overall effects

Question 8

Problems with meta-analysis

Answer 8

Publication bias

“Comparing apples to oranges”

Question 9

Benefits of meta-analysis

Answer 9

Generalisation to broad population possible
Precision and accuracy improve
Differentiate between real and sampling variation