PscanChIP Help

Submitting ChIP-Seq enriched regions

On the right-hand column of the main page several datasets are available for testing the interface. Clicking on any link automatically loads a list of genomic coordinates in the input text box.



For example, click on the NF-YA link. It contains a set of regions of the ENCODE project enriched for the binding of the YA subunit of NF-Y in the K562 cell line. As you can notice, the regions are of 1 bp, matching exactly the point of local maximum (summit) of the peaks. They will be automatically extended by 75 bp in each direction, yielding for the analysis 150 bp regions.
Below the input box, you can choose the source organism (human, in this case), the genome assembly (hg19), the cell line (K562, included in the list) and the matrix database you want to use (Jaspar, Transfac, or you can upload one or more matrices). For the example, all these parameters have been automatically set.
Click "Run!"



In the textbox under the "Run" button, a confirmation message has appeared. Any wrongly formatted region in the input will be reported in the textbox, as well as genomic coordinates not matching the genome assembly selected. Note: a large number of regions on "normal" chromosomes of the assembly reported as wrong and discarded is a likely indicator that you selected the wrong genome/assembly as input.
Results will shortly appear in the middle column of the page (see the

Output:

When you click the "Run" button, a confirmation message (with possible warnings about the input) appears in the text box next to it. The result of the computation will appear in the middle column of the page. The figure shows the output for the NF-Y/K562 example (sample 1).



For each matrix (including those you uploaded, if any), PscanChIP computes an enrichment P-value according to three different criteria. More in detail, from left to right, each line of the output table corresponds to a matrix, and it contains the following information:

1. Name and ID: the name and the database ID of the matrix.
2. Local enrichment p-value (L.PV): describing if the matrix is over- or under-represented in the 150 bp input regions with respect to the genomic regions flanking them. Since the p-value is computed with a two sided t-test, next to the p-value an arrow (L.O/U) indicates whether it means that the matrix motif tends to be over-represented in the sample (red arrow pointing upwards) or vice versa avoided (green arrow pointing downwards).
3. Global enrichment (G.PV), describing if the matrix is over- or under-represented in the input regions with respect to a global background made of a genome-wide collection of putative regulatory (TF accessible) regions in various cell lines. Once again, the p-value is computed with a two sided t-test, next to the p-value an arrow (G.O/U) indicates whether the matrix motif tends to be over-represented in the sample (red arrow pointing upwards) or vice versa avoided (green arrow pointing downwards). This value is computed for user-submitted matrices only if the "Compute global pvalue for additional descriptors" options was set at input.
4. Spearman correlation coefficient (SP.COR): if the input regions were ranked according to their enrichment, this value represents the Spearman rank correlation coeffifient between the ranking of the input regions and their ranking with respect to the motif. In other words, positive values indicate that the more enriched regions are, the best matches they contain for the motif. A correlation around zero denotes no significant correlation, while a negative value is an indicator of anti-correlation (i.e. the more enriched are the regions the worse are motif matches in the regions). If your regions are *not* sorted according to enrichment, this correlation clearly has no meaning.
5. Preferred position (P.POS) and position bias pvalue (P.POS.PV): this indicates the position within the regions where the motif tends to be found more frequently with best matches to the matrix. Coordinates are relative to the center of the regions, which has coordinate 0 . The associated p-value indicates the significance of the positional bias of the motif, computing both according to the number of occurrences of the motif within the region and their score. In a ChIP-Seq experiment, the matrix corresponding to the binding sites of the TF investigated should be found with low p-values at the center of the regions. But, the same holds true (positional bias) also for the TFs binding in association with the main one, as also shown by performing motif-centered analyses.

Performing a motif-centered analysis

A motif centered analysis can be started after a normal analysis, from the detailed output page of the matrix of interest, as explained in the precious section. That is, you have to make Pscan report the occurrences of the motif. Then, you can click on the "Run Pscan-Chip centered on these sites" button.
The program will collect the absolute genomic coordinates of the best match for the matrix in the original regions, and extend them producing 150 regions as with ChIP-Seq summit regions. In this case, thus, coordinate 0 of each region will correspond to the center of the motif selected. A new window will open and the analysis will be started automatically.
The rationale behind this kind of analysis is that, once the binding motif for the TF investigated has been detected, further information can be gathered by re-running the analysis centered on the motif itself, so to discover at a glance whether there exist any significant positional correlation between the motif chosen and the other TRANSFAC, JASPAR or user submitted matrices, and if the correlation is significant from a statistical point of view. In this way, additional candidate TFs binding in a cooperative way (e.g. forming heterodimers) can be discovered.

The output of the motif-centered analysis is identical to the default one, with the sole difference that for the matrix chosen (whose sites will fall exactly in the middle of the regions analysed) PscanChIP will look for the second best match within the regions themselves, in order to detect if there is a preferential arrangements of sites for the same TF, e.g. in case of TFs binding as homodimers. Also, above the Output table, you will find an additional link called Distance hotspots map that opens a graphical representation of the "preferred" positions of the matrices used with respect to the one on which the analysis is centered, that can be browsed (try hovering with the mouse on the red spots in the map).
For example, performing an analysis centered on the NF-Y matrix, you will notice in the result table a large number of different matrices having positional preference (P.POS) not only for the middle of the regions (as for NF-Y itself, since the analysis is exactly centered on its sites, and other factors binding CCAAT-like elements), but also with strong bias and low p-values for different portions, as for example AP1 (FOS+JUN complex), KLF4 (or Sp1), TBP, and E-box motifs (Myc, MAX, Usf1), as shown by their respective occurrence distribution plots (the dotted line correspond to the position of the NF-Y CCAAT motif):

1. NF-YA ChIP-Seq in K562 human cells, assembly hg19. Shows how the binding sites for the factor investigated can be easily found over-represented in the center of the regions, with positive Spearman correlation coefficient with the input regions, which were sorted according to their enrichment. There is a large number of other motifs over-represented, either locally or globally. Performing a motif centered analysis on the NFYA matrix shows several other matrices with a precise positional preference around it (AP1, TBP, SP1, KLF4,E-box motifs like MYC, USF1, MAX, and so on) .
2. 13 "core" TFs in mouse stem cells, assembly mm10. Usually, the binding sites for the factor investigated can be easily found over-represented in the center of the regions (see the article for discussion on E2F1). The global p-value enrichment is able to detect other TFs which have been demonstrated to correlate with the main one in mouse stem cells, while others not correlating have p-values pointing to the fact that their sites seem to be avoided either locally or globally in the regions.
3. Sample 3 - Stat3 in four mouse cell types (sorted). Binding of Stat3 can be inferred in all the experiments, with different co-factors and interactors appearing in each dataset.
4. Sample 4 - E2F1 summits in HeLa-S3 cells. "Canonical" sites for E2F1 appear to be enriched in the dataset, despite other studies suggesting the opposite (lack of sites in E2F1-bound regions). Notice however how E-box binding motifs (like Mycn) virtually present the same type of enrichment, suggesting their possible role in recruiting E2F1, at least in some contexts.
5. NRF1 ChIP-Seq in hESC human cells, assembly hg19. The NRF1 matrix is not included in the TRANSFAC or JASPAR datasets, and it has to be uploaded to the interface. This is what can be done if, for example, a motif finding tool has been run on the regions beforehand, and one or more matrices have been built as a result. Running PscanChIP on them will permit to show their positional bias in the regions, and how they compare and/or correlate with other existing motif descriptors.