SPIN tutorial

1
SPIN tutorial
2
Sorting multidimensionaldata using distance
matrices

• Clustering aims to partition data, such that
  points within a cluster are more similar as
  compared to points outside the cluster.
• Sorting rearranges points into a particular
  one-dimensional permutation, that reflects the
  shape of their arrangement.

3
Shape of cluster
4
Ordering according to ascension
5
Sorting a ring of points
Which is the better approach?
An energy function that penalizes blue points
that are far from the main diagonal.
or
An energy function that penalizes red-points near
the main diagonal.
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
Right
Right
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14

• Sort the genes by setting Width to 10 then
  pressing the Neighborhood button.
• Question 10 How can you identify potential
  "clusters" in the sorted distance matrix?
• Select the last "cluster" in the distance matrix.
  This can be done in several ways
• By dragging a box around it in the distance
  matrix.
• Dragging a box of appropriate height in the
  expression axes.
• Setting the values of the small text boxes below
  the distance-matrix to 377 and 392 respectively.
• Note that the results are generally not very
  sensitive to the exact choice of the "cluster"
• The sorter now highlights the selected region in
  all 3 of the top displays. Zoom in on the
  "cluster" by clicking Zoom in button.
• Press Transpose to view the samples in the space
  of the selected cluster.
• Sort the samples by using Side2side.
• Export the PCA image and add the labels.
• Question 11 What is the connection between the
  labels and the current ordering of the samples?
• Transpose and Zoom out. Look at the ordering of
  the samples (columns) in the expression matrix
  (top middle image).
• Question 12 Why are they ordered in this
  particular fashion?
• Repeat steps 9-11 on the "cluster" of genes
  between 40 and 80.
• Question 13 What is the major partition of the
  samples in these gene-space?
• Question 14 What is the connection to previously
  known labels?
• You may try to zoom in and out of different
  groups of genes and try to discover novel
  partitions of the samples.