Cluster analysis of RNASeq data

Noa · 05-09-2013, 05:37 AM

I have a data set made with 6 developmental time points during development, on a newly assembled transcriptome. We used BLAST2GO to annotate the trasncriptome. In looking for interesting genes or patterns to follow up on, we wanted to perform clustering. Since this is my first move into clustering, does anyone have recommendations? I have no idea how to choose between hierarchial, kmeans, etc (and if kmeans - which k do I choose?)
Any help (even pointing me in the direction of a good intro paper) would be appreciated!
Thanks!
Also- what kinds of normalization do I need to do to preprocess the different stages? I feel that the FPKMs are a bit higher than average in one stage and a bit lower than average in another and have no idea how to deal with this.

chadn737 · 05-09-2013, 06:37 AM

There are pros and cons to each approach. Hierarchical will give you a cluster of clusters. This could be an advantage or completely useless. Hierarchical clustering is also prone to outliers if I remember right. An outlier would form its own cluster entirely. On the other hand k-means requires you to define the number of clusters. This can be done by using the Gap statistic of Tibshirani.

priya · 06-13-2013, 08:31 AM

Hi, Hierachical clustering is a nice way of representing the samples difference and to look at the relationship between the samples in the initial stages of how raw data look like. Although the way the clusters are formed in tree corresponds to how we calculate the distance measure(Single/Wards/Complete/Average) and the type of method (Euclidean/Pearson)used to calculate the distances between two data points . I have 16 RNA-seq samples, tried to perform hierarchical clustering on dataset, by using Euclidean distance measure and Wards methods, the tree genertated was different if i use Single-linkage method. Although by Single linkage method the tree is making sense in terms of biology, but I am not completely clear of the point which will be the optimal method to consider when we look at the RNA-seq data having count values??

Any suggestions please??

dietmar13 · 06-13-2013, 10:41 PM

non-negative matrix factorization (NMF) also allows unsupervised clustering, defining also the number of clusters which represents the data best using the cophenetic correlation index. in my hands this method produces more robust clusters compared to hierarchical clustering...

dietmar

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
RNAseq data analysis	chknbio	Bioinformatics	8	05-25-2012 12:44 AM
Basic steps in prokaryotic RNAseq data analysis	Muhidini	Bioinformatics	0	05-09-2012 07:27 AM
[NGS - analysis of gene expression data] Machine Learning + RNAseq data	Chuckytah	Bioinformatics	7	03-05-2012 04:16 AM
RNAseq - low cluster density - possible inhibitor?	OnUrMark	Sample Prep / Library Generation	11	05-07-2011 08:05 PM
Companies offering NGS data analysis including RNAseq	agseq	Bioinformatics	0	05-05-2011 07:24 AM

05-09-2013, 05:37 AM	#1
Noa Member Location: haifa israel Join Date: Jun 2011 Posts: 62	Cluster analysis of RNASeq data I have a data set made with 6 developmental time points during development, on a newly assembled transcriptome. We used BLAST2GO to annotate the trasncriptome. In looking for interesting genes or patterns to follow up on, we wanted to perform clustering. Since this is my first move into clustering, does anyone have recommendations? I have no idea how to choose between hierarchial, kmeans, etc (and if kmeans - which k do I choose?) Any help (even pointing me in the direction of a good intro paper) would be appreciated! Thanks! Also- what kinds of normalization do I need to do to preprocess the different stages? I feel that the FPKMs are a bit higher than average in one stage and a bit lower than average in another and have no idea how to deal with this.

06-13-2013, 10:41 PM	#4
dietmar13 Senior Member Location: Vienna Join Date: Mar 2010 Posts: 107	non-negative matrix factorization (NMF) non-negative matrix factorization (NMF) also allows unsupervised clustering, defining also the number of clusters which represents the data best using the cophenetic correlation index. in my hands this method produces more robust clusters compared to hierarchical clustering... dietmar

05-09-2013, 06:37 AM	#2
chadn737 Senior Member Location: US Join Date: Jan 2009 Posts: 392	There are pros and cons to each approach. Hierarchical will give you a cluster of clusters. This could be an advantage or completely useless. Hierarchical clustering is also prone to outliers if I remember right. An outlier would form its own cluster entirely. On the other hand k-means requires you to define the number of clusters. This can be done by using the Gap statistic of Tibshirani.

06-13-2013, 08:31 AM	#3
priya Member Location: sweden Join Date: Apr 2013 Posts: 57	Hi, Hierachical clustering is a nice way of representing the samples difference and to look at the relationship between the samples in the initial stages of how raw data look like. Although the way the clusters are formed in tree corresponds to how we calculate the distance measure(Single/Wards/Complete/Average) and the type of method (Euclidean/Pearson)used to calculate the distances between two data points . I have 16 RNA-seq samples, tried to perform hierarchical clustering on dataset, by using Euclidean distance measure and Wards methods, the tree genertated was different if i use Single-linkage method. Although by Single linkage method the tree is making sense in terms of biology, but I am not completely clear of the point which will be the optimal method to consider when we look at the RNA-seq data having count values?? Any suggestions please??