MicroExons KO
1 Setup
1.1 Gene Annotations
1.2 Build categories DB
We are using the databases of shinygo, which are based on Ensembl Release 104.
2 Samples
3 PCA
Let’s do a first PCA to explore the samples
And evaluate what factors correlate with the PC.
There is a clear Concentration and Round “effect” on the samples.
4 Batch and Concentration Correction
4.1 Select Genes for batch correction
Genes were selected for further analysis if in two or more samples from the same founder the gene had at least 10 read count.
4.2 Empirical Bayes-moderated adjustment for unwanted covariates
This method removes variation due to unwanted covariates, while preserving variation related to covariates of interest, in our case the unwanted covariates are batch and concentration and the ones of interest are founder and genotype. The Empircal bayes-moderated linear regression was calculated based only on the WT samples and applied to all. The way of removing these effects is to first normalize (log scale) the counts and for that we used the variance stabilizing transformation function from DESeq2. The design of the experiment was just the intercept (~ 1).
It is important to mention that for some samples the concentration value is unclear and therefore are removed from further analysis, but can be added if the concentration is clarify. These samples are:
- Embr_5dpf_Shank3b_1_DEL_d
- Embr_5dpf_Shank3b_1_WT_d
- Embr_5dpf_Shank3b_2_DEL_d
- Embr_5dpf_Shank3b_2_WT_d
- Embr_5dpf_Vdac3_5_DEL_bd
- Embr_5dpf_Vdac3_5_WT_bd
This is how the expression distributions look like for each sample
4.3 New PCA
This is how the new PCA looks like after the Batch and Concentration correction:
Neither round or concentration are significant any more in the firsts PC, thus meaning that the correction has been applied correctly.
It is important to notice that there is not clear division between DEL and WT
5 Differences between DEL and WT
p-values and proper statistics were not made because of the lacking of replicates in most deleted MICs.
Since corrected values are log-scale we could just calculate the difference between DEL MIC and WT per founder.This is how the distributions of the differences look like:
As expected most genes do not show differences between DEL and WT. Nevertheless, we can identify if a certain group of genes show differences among multiple of the comparisons. For that we sum the absolute difference and performed a GSEA (only in positives values) using KEGG, GO, Biological Processes, Molecular Functions, Cellular Compartment and Diseases gene annotations.
5.1 General GSEA
|
| | 0%
|
|================== | 25%
|
|=================================== | 50%
|
|==================================================== | 75%
|
|======================================================================| 100%
|
| | 0%
|
|= | 2%
|
|== | 3%
|
|=== | 5%
|
|===== | 7%
|
|====== | 8%
|
|======= | 10%
|
|======== | 11%
|
|========= | 13%
|
|========== | 15%
|
|=========== | 16%
|
|============= | 18%
|
|============== | 20%
|
|=============== | 21%
|
|================ | 23%
|
|================= | 25%
|
|================== | 26%
|
|==================== | 28%
|
|===================== | 30%
|
|====================== | 31%
|
|======================= | 33%
|
|======================== | 34%
|
|========================= | 36%
|
|========================== | 38%
|
|============================ | 39%
|
|============================= | 41%
|
|============================== | 43%
|
|=============================== | 44%
|
|================================ | 46%
|
|================================= | 48%
|
|================================== | 49%
|
|==================================== | 51%
|
|===================================== | 52%
|
|====================================== | 54%
|
|======================================= | 56%
|
|======================================== | 57%
|
|========================================= | 59%
|
|========================================== | 61%
|
|============================================ | 62%
|
|============================================= | 64%
|
|============================================== | 66%
|
|=============================================== | 67%
|
|================================================ | 69%
|
|================================================= | 70%
|
|================================================== | 72%
|
|==================================================== | 74%
|
|===================================================== | 75%
|
|====================================================== | 77%
|
|======================================================= | 79%
|
|======================================================== | 80%
|
|========================================================= | 82%
|
|=========================================================== | 84%
|
|============================================================ | 85%
|
|============================================================= | 87%
|
|============================================================== | 89%
|
|=============================================================== | 90%
|
|================================================================ | 92%
|
|================================================================= | 93%
|
|=================================================================== | 95%
|
|==================================================================== | 97%
|
|===================================================================== | 98%
|
|======================================================================| 100%
|
| | 0%
|
|== | 2%
|
|=== | 4%
|
|===== | 7%
|
|====== | 9%
|
|======== | 11%
|
|========= | 13%
|
|=========== | 16%
|
|============ | 18%
|
|============== | 20%
|
|================ | 22%
|
|================= | 24%
|
|=================== | 27%
|
|==================== | 29%
|
|====================== | 31%
|
|======================= | 33%
|
|========================= | 36%
|
|========================== | 38%
|
|============================ | 40%
|
|============================== | 42%
|
|=============================== | 44%
|
|================================= | 47%
|
|================================== | 49%
|
|==================================== | 51%
|
|===================================== | 53%
|
|======================================= | 56%
|
|======================================== | 58%
|
|========================================== | 60%
|
|============================================ | 62%
|
|============================================= | 64%
|
|=============================================== | 67%
|
|================================================ | 69%
|
|================================================== | 71%
|
|=================================================== | 73%
|
|===================================================== | 76%
|
|====================================================== | 78%
|
|======================================================== | 80%
|
|========================================================== | 82%
|
|=========================================================== | 84%
|
|============================================================= | 87%
|
|============================================================== | 89%
|
|================================================================ | 91%
|
|================================================================= | 93%
|
|=================================================================== | 96%
|
|==================================================================== | 98%
|
|======================================================================| 100%
|
| | 0%
|
|== | 3%
|
|==== | 6%
|
|====== | 8%
|
|======== | 11%
|
|========== | 14%
|
|============ | 17%
|
|============== | 19%
|
|================ | 22%
|
|================== | 25%
|
|=================== | 28%
|
|===================== | 31%
|
|======================= | 33%
|
|========================= | 36%
|
|=========================== | 39%
|
|============================= | 42%
|
|=============================== | 44%
|
|================================= | 47%
|
|=================================== | 50%
|
|===================================== | 53%
|
|======================================= | 56%
|
|========================================= | 58%
|
|=========================================== | 61%
|
|============================================= | 64%
|
|=============================================== | 67%
|
|================================================= | 69%
|
|=================================================== | 72%
|
|==================================================== | 75%
|
|====================================================== | 78%
|
|======================================================== | 81%
|
|========================================================== | 83%
|
|============================================================ | 86%
|
|============================================================== | 89%
|
|================================================================ | 92%
|
|================================================================== | 94%
|
|==================================================================== | 97%
|
|======================================================================| 100%
|
| | 0%
|
|====== | 9%
|
|============= | 18%
|
|=================== | 27%
|
|========================= | 36%
|
|================================ | 45%
|
|====================================== | 55%
|
|============================================= | 64%
|
|=================================================== | 73%
|
|========================================================= | 82%
|
|================================================================ | 91%
|
|======================================================================| 100%Dotplots are caped on the y axis, meaning that higher -log10(padj) than 5 are set 5. KEGG pathways and Diseases do not have a table since results are well summarized in the figure.
5.2 MIC-DEL GSEA
We performed a GSEA on different annotations (KEGG and Disease) for each comparison between DEL and WT and selected only those pathways that either showed significance (adjusted p-value < 0.01) in at least 2 deleted MIC replicates (with same sign of NES) or in at least half of the deleted MICs without replicates.
Once I heard in a conference that this type of analysis most of the time lead to you to a dead-end or to categories that are too general, but sometimes you get an incredible result and you need to dig deeper. This later seems our case, the database I started with consist of more than 900 diseases and finding autistic related terms is not casualty and too perfect. Looking deeper into the two categories the size of the terms is 125 for Autistic disorder and 72 for Autism spectrum disorder, and they only intersect in 23 genes. This means that either those 23 genes are driving the significance or there is some convergence. It is also intriguingly the opposite enrichment score on others diseases.
According to the results shown in both GSEA the samples can be divided into two groups. For that we did a hierarchical clustering using as distance 1 - pearson correlation between delta corrected VST differences (DEL - WT). This is the out-coming tree:
Sample order will be maintain in all heatmaps.
I know there has been some concerns about the initial concentration and the results, therefore I will next try to explore that relation further. Lets first see if NES correlates (spearman) with the ratio log10(Concentration DEL / Concentration WT):
It doesn’t seem to correlate, but is there any differences in the ratio between positive and negative NES, for each of the significant categories?
Or only the concentration of DEL?
Apparently concentration does not seem to have an effect con this result. In following sections I will try different ways to keep validating this.
5.2.1 Intersected genes of autistic terms
5.2.2 Unique genes for “Autistic disorder” term
5.2.3 Unique genes for “Autism spectrum disorder” term
The trends are clear between the two categories, not only in the intersected genes but also in the unique for each term.
6 WT Randomization
One of the possible controls is to randomize the WT samples, this means that founders will not match. For this approach I performed 10 randomization and tested GSEA for diseases.
The following heatmap shows the result of the simulations. A disease was considered enriched in each randomization the same manner as in the previous GSEA analysis. Here are shown pathways which showed significance more than 5 times in a single sample. NES are the average of the 10 randomization and alpha depicts the number of times the term showed significance per sample.
Besides autistic disorder is one of the terms that survived the thresholds the signal out of this is less clean. This could mean that the founder pairing might be substituted, therefore I compared each DEL vs the mean of WT.
7 Mean WT
If founder might not play a crucial role what would happen if each DEL is compared with the mean of WTs.
Autism results are still there, the counterpart diseases are also there and the two MIC groups maintain on the big scale, this means that for further analysis the mean of WTs could be used for comparisons.
8 Ranked
Another way to test if concentration has a clear effect on the disease result is to rank WT and DEL samples based on concentration and match samples based on those ranks. Meaning that we will compare the WT with most concentration vs DEL with most concentration (Top-Top) or the other way around, most concentration vs least one (Top-Bottom). The hypothesis, if concentration is playing a role, is that in the Top-Bottom the signal of autism pathway is more clear.
8.1 Top-Top
8.2 Top-Bottom
After these rankings it is still not clear if the concentration is involved in the “Autism” result, since it keeps coming out.
9 Without Batch&Concentration correction
How would it be without correcting the “unwanted” variation?
10 Comparing WT
If the autism pathway is solely dependent of concentration when comparing Top-Bottom ranked WTs it should show significance.
Fortunately, the autistic term did not came out significant in here. It is important to mention that concentration is not playing a role anymore in these samples since bigger concentration differences do not show more significant pathways. Another important result out of this is that some WT samples look problematic, in the extremes are WT samples ranked 3 or 27 and 6 or 24, and more mild 11|19, 16|14 and 10|20. Further analysis can be done here.
11 Grouping
With the original results, when comparing between founders DEL vs WT, we identified two groups of MIC deletions. In order to test the grouping, I performed similar clustering from the above mentioned controls. Groups are standardized based on “Evi5b_3_b”, meaning that Evi5b_3_b will be always in the same group. This is because in the randomization section this sample showed always significance for autistic terms.
This is the result for only deleted MICs with any type of replicates:
This shows that for most DEL-MICs the grouping is consistent despite the starting comparison.
11.1 Grouping in PCA
Lets do a PCA only in the DEL samples (using the adjusted counts) and identify the groupings. (Groups are not standardized here)
12 Ribosomal Genes
During this analysis there is a term, besides autism, that keeps popping up in the disease GSEA and this term is Diamond-Blackfan Anemia. This disease is tied linked with abnormal ribosomal protein genes. Ribosomal related terms come significant in all the top terms of the other analyses. Therefore in this section I explore this part a little bit more.
This is a correlation plot of all ribosomal proteins based on the adjusted VST values (ignore the first plot):
Based on this two groups or ribosomal proteins were considered and for each sample the proportion of reads mapped to each ribosomal group was calculated. These values are then added to the metadata of the PCA after the correction
And the results show that PC correlates with both Ribo proteins groups.
Then I used the empircal bayes linear regression to eliminate these factors. The idea behind is that all samples should have the same proportion of ribosomal proteins.
This might not be biological true.
And then if we look for significant diseases we don’t get the autistic result anymore, neither the two groups.
And KEGG results look like:
