15 In Silico Prediction of the Point of Departure (POD) …
305
PODs using the most sensitive pathway approaches. However, different choices can
be made for both experimental and modeling aspects for obtaining transcriptional
PODs. Researchers should consider carefully how to proceed with these types of
studies based on a number of considerations. In the next subsection, we will summarize some important issues to consider when inferring PODs from transcriptomic
data.
15.2.2 Important Issues for Inferring PODs
from Transcriptomic Data
After the work of Thomas et al. [16], a number of researchers have proposed variations
to their approach. Though some systematic comparisons have been carried out and
recommendations are given (e.g., [20]), a consensus has not been reached on all
issues. In this review, we do not try to give comprehensive guidance for every question
that might arise from a study. Rather, we will present a list of important issues and
possible choices for solutions. Fortunately, experience so far in this field suggests
that reasonable results can often be reached via different variations in approaches.
Experiment designs and technology considerations. Principles for designing dose
response experiments aimed at determining PODs are well known. Though multiple
approaches exist for determining PODs [e.g., lowest-observed-adverse-effect levels (LOAELs), no-observed-adverse-effect-levels (NOAELs)], generally the BMD
approach is used for transcriptional PODs. The advantage of the BMD approach is
that it can utilize all data points to obtain more stable results [21]. To accurately
model the BMD, it is ideal to have doses covering the whole range of the dose
response curve. However, since there are thousands of genes, it is impractical to
expect the doses are ideal for modeling all genes. Usually, the doses are chosen
based on knowledge about apical endpoints.
Besides doses, studies reported so far often include samples taken at multiple time
points after the dosing started. This helps shed light on whether the transcriptional
pattern changes with time. Reported studies show that the transcriptional PODs tend
to be relatively stable even if the most sensitive pathways are not in a duration of
several weeks. Considering the high cost associated with long-dosing periods, it is
thus reasonable to focus on short-dosing periods (in a matter of days) when budget
is limited. Due to cost considerations, the number of animals evaluated per dose
per time point for transcription is often small (as low as three). It is suggested that
more emphasis on the number of doses over the sample size at each dose might be
preferable [22] when operating under budget constraints.
Another important factor to consider is technology. Though most studies published
so far have used microarrays (usually from Affymetrix), RNAseq is expected to
become more popular due to decreased cost and the improvement in quality. In the
broad area of transcriptional profiling, RNAseq has been shown to provide higher
precision, a great dynamic range, and the ability to detect novel transcripts. But
305
PODs using the most sensitive pathway approaches. However, different choices can
be made for both experimental and modeling aspects for obtaining transcriptional
PODs. Researchers should consider carefully how to proceed with these types of
studies based on a number of considerations. In the next subsection, we will summarize some important issues to consider when inferring PODs from transcriptomic
data.
15.2.2 Important Issues for Inferring PODs
from Transcriptomic Data
After the work of Thomas et al. [16], a number of researchers have proposed variations
to their approach. Though some systematic comparisons have been carried out and
recommendations are given (e.g., [20]), a consensus has not been reached on all
issues. In this review, we do not try to give comprehensive guidance for every question
that might arise from a study. Rather, we will present a list of important issues and
possible choices for solutions. Fortunately, experience so far in this field suggests
that reasonable results can often be reached via different variations in approaches.
Experiment designs and technology considerations. Principles for designing dose
response experiments aimed at determining PODs are well known. Though multiple
approaches exist for determining PODs [e.g., lowest-observed-adverse-effect levels (LOAELs), no-observed-adverse-effect-levels (NOAELs)], generally the BMD
approach is used for transcriptional PODs. The advantage of the BMD approach is
that it can utilize all data points to obtain more stable results [21]. To accurately
model the BMD, it is ideal to have doses covering the whole range of the dose
response curve. However, since there are thousands of genes, it is impractical to
expect the doses are ideal for modeling all genes. Usually, the doses are chosen
based on knowledge about apical endpoints.
Besides doses, studies reported so far often include samples taken at multiple time
points after the dosing started. This helps shed light on whether the transcriptional
pattern changes with time. Reported studies show that the transcriptional PODs tend
to be relatively stable even if the most sensitive pathways are not in a duration of
several weeks. Considering the high cost associated with long-dosing periods, it is
thus reasonable to focus on short-dosing periods (in a matter of days) when budget
is limited. Due to cost considerations, the number of animals evaluated per dose
per time point for transcription is often small (as low as three). It is suggested that
more emphasis on the number of doses over the sample size at each dose might be
preferable [22] when operating under budget constraints.
Another important factor to consider is technology. Though most studies published
so far have used microarrays (usually from Affymetrix), RNAseq is expected to
become more popular due to decreased cost and the improvement in quality. In the
broad area of transcriptional profiling, RNAseq has been shown to provide higher
precision, a great dynamic range, and the ability to detect novel transcripts. But