Data import
-----------
To analyze a microarray dataset one needs both the CEL files
and platform data (CDF,GIN,sequence,probe sequence). These can
be downloaded from Affymetrix. Functions to read and combine all
these files can be found in the import directory.
1) cdf = mt_readcdf('path/cdffile')
2) gin = mt_readgin('path/ginfile')
3) pa = mt_readprobe_annot('path/probe_sequencefile') [optional, necessary for background/hybridization correction]
4) seq = mt_readseq('path/seqfile'); seq = mt_read_seq('path/anotherseqfile',seq); [optional, only necessary for amplification correction]
5) celpaths = dir_filter('path/','.cel'); cel = mt_readcel(celpaths);
5) probes = mt_cel2probes(cel,cdf,gin,pa,seq);
6) probes = mt_readgene_annot(probes, 'path/annotfile.csv') [optional, adds probeset annotation from affymetrix reanalysis]
Normalization
-------------
Then to normalize, use:
probes = mt_normalize(probes);
This script performs the different normalization steps in order. If
one needs more control, one can perform each step separately:
1) <dataset> = mt_bg_est(<dataset>,'m_estimation','reg_max',10)
Estimate optical and background signal.
2) <dataset> = mt_cor_hybamp(<dataset>, 'm_estimation','use_amplification')
Estimate sequence/amplification model parameters for each array
3) <dataset> = mt_cor_qq(<dataset>)
(optional) Perform quantile quantile normalization
3) <dataset> = mt_cor_image(<dataset>,9,'use_optical')
Correct for array location effects
4) <dataset> = mt_cor_qq(<dataset>)
Redetermine quantile quantile normalization factors
5) <dataset> = mt_sum_plm(<dataset>,'m_estimation')
Calculate summarized data.
Affy RDN interface
------------------
An alternative interface is available through the affy_rdn script developed by Michel Bellis.
This interface is also suitable for use from the command line.
#load cel files and save them in rdn format
affy_rdn('cel', 'experiment_name', 'path_to_dir/with/cel_files')
#load chip info files and save them in rdn format
affy_rdn('info', 'experiment_name', 'path_to_dir/with/cel_files', 'chip_name', 'path_to_dir/with/chip_info_files')
#run rdn normalization, and save results
affy_rdn('rdn', 'experiment_name', 'path_to_dir/with/cel_files', 'chip_name', 'path_to_dir/with/chip_info_files')
#print results
affy_rdn('print', 'experiment_name', 'path_to_dir/with/cel_files')
For more information, see the documentation in the affy_rdn file.
Differential expression detection
---------------------------------
To determine statistics for differential expressed genes, one can use
mt_sig and mt_sig_perm in combination with various stat functions.
Annotation labels can be stored in an annot cell structure, where
each cell is an annotation vector/matrix.
Annot cell format:
1. Continous labels: vector. Empty/non-measured: NaN
2. 2-class: matrix with shape (narray,2), where 0/1 indicates class membership
3. 2-class + extra permutation samples: matrix with shape (narray,3), where
extra permutation samples are indicated in third column
4. non-numeric labels (e.g. for plotting): use cell array of strings
Example:
<dataset>.annots = {
{'Sample 1', 'Sample 2', 'Sample 3', 'Sample 4'},
[1,2,3,4],
[0,1;1,0;0,0;1,0];
}
Release history
--------------
0.10: Initial release
0.11: solve bug in mt_sel_array and mt_sel_genes
0.12: gene_seq field was incorrectly stored as gene_sequence by mt_cel2probes
0.13: - readseq now does not add probesets for which there is already a sequence.
- add rmoutlier code for mt_cel2probes to distribution
- solve bug in check code of mt_cel2probes
0.14: - mt_cel2probes did not attach sequence data if
no rmoutliers parameter was given
0.15 - improved progress report
- solve bug in mt_sel_genes
- add support for pm-only arrays
- remove assumption of 25-length probe sequences
- fix reference to gene_seq field
- add support for new binary cel file format
- add support for new binary cdf file format
- improve gene name mapping in mt_readgene_annot
- make mt_readseq a bit more robust against empty lines
0.20 - detection of array size did not work correctly for all array formats (fixed, thanks to Michel Bellis)
- reduce cel struct, cdf struct and probes structure memory usage, precision of
various data structures reduced to single precision (thanks to Michel Bellis)
- removed unnecessary caching of weighted variable matrices in mt_bg_est and mt_cor_hybamp (thanks to Michel Bellis)
- outlier removal option in mt_cel2probes did not work anymore, fixed
- reduce_probes option for mt_bg_est and mt_cor_hybamp (increases speed, reduces memory at cost of some precision)
- mt_sum_plm memory usage reduced, some optimizations
- add mt_normalize function
- add mt_plot_array function
- add mt_plot_gene function
- add mt_plot_seq_effects function
- add mt_plot_amp_effects function
- reorganization of directory structure. No addpath command necessary anymore.
- add import documentation
- add mt_sig_perm function (permutation testing)
- add some stat functions (mt_stat_[sam,fc,f,cor,rankcor,ranksum])
0.21 - allow points in probeset names (thanks to Michel Bellis)
- add missing mt_cummin function
- add documentation to mt_sig_perm and on annot format
- add mt_sig function, to calculate a certain statistic over genes
0.22 - added export/import function for R expression data files
0.23 - cel file header corruption check
- handle extra columns in probe annot file
- make buffer large enough for very large R expression files
- mt_real_signal non private
- selection of indmm now backwards compatible
- option for label ordering mt_plot_gene
- accurately determine maximum number of permutations in multi-class case
- plotting of amplification effect without position info, only use probesets with more than 5 probes
- auc stat function
- plot array correlation, signal density
- plot averaged local chip deviation of arrays
- add background per probeset output to mt_sum_plm
0.24 - added affy_rdn.m command line script developed by Michel Bellis
- added an mt_expr_signal function to recover the normalized summarized signal