- TF activity timing
- TF activity score threshold
- Randomization error
- TFInput warning
- Path statistics
- Disconnected network
Q: How do I identify the time points at which TFs are first active?
A: Visually, the Key Transcription Factors interface in the DREM display can be used to show TFs the first time they are active on a DREM path based on the SDREM activity score. This information is not directly saved to a text file, but it can be reconstructed from the N.model.activitiesDynamic file that SDREM produces.
N.model.activitiesDynamic, where N is the final iteration of SDREM, is a tab-delimited text file that contains three columns:
- TF identifier
- TF activity score at a particular split node in the DREM paths
- Tree depth of that split node
The tree depth of the split node can be used to identify the first time a TF is active. The depth is 0-based so if you have time points of 0, 5, 15, 30, 60 minutes the corresponding tree depths are 0, 1, 2, 3, and 4. By parsing or manually examining N.model.activitiesDynamic, you can identify the first time (minimum tree depth) at which a TF exceeds the activity score threshold you set in the Key Transcription Factors slider. Note that for activity scores, the slider is used to threshold TFs that have a score >= 10^X.
To illustrate, suppose you set X = 4.5 in the slider. Then a TF will be shown the first time its activity exceeds 10^4.5 = 31622 on a DREM path. You can use N.model.activitiesDynamic to filter out the rows where the activity score is < 31622. This will give you a list of TF identifiers and tree depths of the split nodes where those TFs attained the high activity score. You can sort by TF identifiers to determine the smallest tree depth with a high activity score for each TF. This gives you an index, which can be used to recover the time point. Using the example time points from above, if a TF exceeds the activity score threshold at split nodes with depth 2, 3, and 3, you would take 2 as the minimum depth (time point index), which corresponds to the 15 minute time point.
The N.model.activitiesDynamic file can also be used to identify all time points at which a TF is active. In the example above, the TF is active at split nodes with depth 2 and 3 meaning it is active at 15 and 30 minutes.
Q: How do I choose a TF activity score threshold when visualizing active TFs in the DREM display?
A: The file N.targetsStd, where N is the final iteration of SDREM, can be used to obtain to an activity score threshold that will display all of the TFs that were designated as targets in the SDREM network orientation stage. The last column in this tab-delimited text file gives the maximum activity score for each active TF. Therefore, you can find the minimum of these values and use it (or a value slightly less than it) as the threshold. Take log10 of this activity score to obtain the value of X to set in the DREM Key Transcription Factors slider.
Q: Why does SDREM throw a randomization error in alg.DREMInterface.randomizeBindingPriors?
A: This can occur when the TF-gene interactions file has multiple rows for the same gene. SDREM assumes that each TF only appears in one column and each gene only appears in one row.
Q: What does the warning The directory TFInput was not found.Directory not found mean?
A: This warning comes from the orignal version of DREM and its graphical interface.
It can be safely ignored.
TFInput is a hard-coded directory name that DREM uses to populate a list of TF-gene interaction files in a drop-down menu.
SDREM does not use a user-selected TF-gene interaction file from this menu.
The preferred way to specify TF-gene interactions is through the binding.priors.file SDREM property.
To remove the warning, create a directory named TFInput.
Q: What are the stderr warnings about updating path statistics?
A: Warnings like
97 ms to update path stats
Updating path statistics
are part of the normal logging. This is written to stderr but does not indicate a problem.
Q: Why are some of the nodes in the topPathEdges_itr10.sif network disconnected from others?
A: By default, topPathEdges_itr10.sif filters many nodes and edges to produce a simpler, more interpretable visualization of the most important nodes and edges in the SDREM network.
Edges that are not members of a top-scoring path are excluded.
Edges that do not have a Source, Target, or Internal node for both endpoints are excluded.
Internal nodes are nodes that have at least 1% of the top-scoring paths pass through them.
The 1% threshold is controlled by node.thresh in the SDREM properites file.
The complete network is available in pathEdges_itr10.txt.