|
HOME | DESIGN | MICROARRAY SAMPLES | NORMALIZATION | SIGNIFICANCE TESTING | SAM | H. CLUSTERING | PAM CLUSTERING | PAM CLASSIFICATION | CONCLUSIONS |
This web page was produced as an assignment for a course on Statistical
Analysis of Microarray Data at
Experimental design:
Tissues samples were taken from frozen surgical specimens at
miRNA microarrays were printed at the Standfor functional genomics facility.
The arrays had a total of 668 probes spotted in duplicate. The probes include
328 known human miRNAs, 113 mouse miRNAs, 45 rat miRNAs, 154 predicted human
miRNAs and 28 control probes.
Dye assignments for the microarrays were not readily available. However, the
fluorescence ratio used for analysis was described as
"sample/reference", which indicates that a dye-swap was not used.
Selection criteria (described here under "Data
Analysis") limited them to 87 miRNAs (out of 668 probes) for analysis.
Once candidate miRNAs were identified, they were analyzed using unsupervised
hierarchical clustering analysis and SAM. By doing so, they identified miRNAs
that were most associated with certain tumor types.
In order to further confirm and extend the miRNA library being used, the
researchers supplemented the microarray data by cloning and sequencing small
RNA libraries from 10 of the sarcomas and 2 normal skeletal muscle samples. The
biggest advantage to doing this was finding novel miRNAs that were not included
in the aforementioned chip. This allows for a more complete miRNA profile to be
constructed for each of the tumor types.
Possible Improvements:
The biggest concern in their experimental design is the use of references,
which was not well described. Because they had two types of references, normal
smooth muscle tissue and skeletal muscle tissue, from 5 and 2 human sources
(respectively), there is no standard procedure for dealing with the references.
Furthermore, although it wasn't described, it appears as though a dye-swap was
not performed. This could be a concern because of dye-bias.
Finally, though they ran "duplicates", they were on the same chip.
From our reference chip picture here, it appears
that the references aren't directly next to each other, which is a plus.
However, ideally when things are run in duplicate, they are run on separate
physical chips rather than on the same chip in different places.
My suggestion for an improved experimental design would be a balanced design as
follows: figure out which sarcomas need which types of references (normal vs.
skeletal) and count each of these up. Suppose that 12 tumor samples need
skeletal references, and 15 need normal references. We would need, then, 12
independent skeletal samples and 15 independent normal samples. Using a chance
device, choose 6 skeletal tumors and pair with 6 skeletal references. Dye the
tumor RNA red and the reference green. Do the same with 7 normal tumors. Repeat
the procedure, this time dying the tumor RNA green, and using 8 normal tumors
and references instead of 7.
This would be my best guess at what a balanced design looks like for this
experiment.