February 9, 2016
Classwork
Clustering Activity
Clustering
Clustering: grouping genes and samples together and presenting them in a specific order.
Clustering is used as a data reduction analysis. It is representative of data points rather than an entire data set. When clustering, we seek to gain an understanding of patterns in a data set, so that they may be tested statistically. While analyzing patterns, it is important to consider the utility of log transformations, co-regulations, and direct/indirect relationships of genes. Both negative and positive correlations can be interesting and lead to important discoveries.
Hierarchical Clustering: joins the two most similar genes, then the next two most similar genes or cluster of genes until all genes have been joined.
In hierarchical clustering, after two genes or cluster of genes are joined, they cannot be pulled apart regardless of what future discoveries in data reveal. The biggest problem with hierarchical clustering is that it does not consider all data components together. Furthermore, no gene is left behind in hierarchical clustering; correlations begin with a value of 1 and end with a value of -1.
K-means Clustering: specifies how many clusters to form by randomly assigning each gene to one of k different clusters.
In K-means clustering, the average expression of all genes in each cluster is used to create k pseudo genes. Genes can be rearranged by assigning each one to the cluster represented by the pseudo gene to which it is most similar. K-means clustering can be repeated until there is convergence.
Supervised Clustering: finds genes in expression file whose patterns are highly similar to the desired gene or pattern.
Supervised clustering adds the closest gene first. Then, the gene closest to all of the genes already in a cluster is added. This process continues as long as the added gene is within the specified distance of genes already in cluster. The specified distance from one gene to a set of genes can be defined as the maximum, minimum, or average of all distances to individual members of the set (complete, single, and average linkage, respectively).
Cutting the Tree: the process of grouping genes by determining a threshold value in the dendrogram.
In cutting the tree, cut the dendrogram at different points and see what genes or clusters of genes are still clustered together. Genes that are still together are part of a cluster. Different clusters arise depending on where the tree was cut.
Intensity plots compare gene expression profiles.
Intensity plots
Comparing gene expression profiles, or guilt by association Proximity measures: correlation, Euclidean distance, inner product XY, Hamming distance, L1 distance, dissimilarities ma or may not be metrics (triangle inequality, looseley referred to as distance) WE WANT TO COMPARE GENES AND EXPRESSION PATTERNS BETWEEN FED AND NON-FED. How do you compare one thing to a group of things? How do you measure similarity/dissimilarity to the cluster? -Define cluster (take averages of things belonging to cluster) and then treat it like an individual to compare to other things. OR average all the distances. OR do max and min (it's a part of the cluster if it is closest to one of them or close to all of them) All of these are linkage methods (Complete linkage, incomplete linkage, mediode, etc.)
Are there things you can cluster where you know the number??
TRACK GENES THAT MATCH WITH A TRANSCRITPION FACTOR- Transcription factor might be small, but we want to see what has big changes that correlate with that.
Use QT Clust instead of heat map: MAIN IDEA: 1) each gene builds a supervised cluster, 2) Gene with "best" list, and genes in its list, becomes next cluster, 3) remove these genes from consideration, and repeat, 4) stop when all genes are clustered, or largest cluster is smaller than user specified threshold.
Gene with the biggest numbers/most genes is the group that we are looking at. We are calling it a cluster, now those genes are not part of anyone's group. Now look for next biggest group and get a different cluster. THERE IS NO ONE PERFECT, CORRECT ANSWER. LOOK FOR THINGS THAT MEAN SOMETHING TO YOU. Chase things you are interested in them, look for things that are similar, and then keep pulling things into your group. PRACTICE RESTRAINT.
It would help to have gene ontology terms to help with clustering. Cluster transcription factors and look at those.
