I am currently using TCGA data available in cbioportal to perform analysis of different p53 targets. There's mRNA level data of all patients and all genes and about half of the patients have protein levels information for just some proteins. Both mRNA levels and protein levels are given in Z-scores. Since I was just getting started with the analysis and still becoming familiarized with everything, I decided to try and plot the p53 protein z-score vs the p21 mRNA z-score to see what it looked like, since p21 is a well-known p53 target. But to my surprise, the scatter plot looked quite random and uncorrelated (I attached below the image of the plot I did in R). I performed a correlation test between the p53 protein z-score and p21 mRNA z-score using Spearman's coefficient (since the distributions aren't normal) and got a very low number, around -0.002. I wanted to know if there's an error in the statistical methodology I'm applying and if not, what could be the explanation to this, since p53 has been widely proven to activate p21 so you would expect at least *some* degree of correlation.
Thanks in advance and sorry if it's a dumb question, I'm new to the field of databases and biostatisical analysis.
This is a common observation. Some possible reasons:
1. Gene targets vary among cancer types.
2. strong correlations like we expect (r>0.7) are rarely achieved in these datasets.
3. Total protein levels are rarely proportional to mRNA levels.
4.. Tumor samples are actually not pure (contain a mixture of cells, including immune cells). You can check our publication o TOX protein, where extensive correlation analysis has been done.
5. A verifying approach can be you take that correlation data from cBioPortal, do a GSEA, and check whether p53 protein levels are associated with overall p53 pathway activation ( target genes).
6. Intratumor heterogeneity (not all cells show activation of processes as we achieve in cell culture plates). A classical example is LIG4 (DNA ligase 4), which we have recently shown (along with previous reports) to be downregulated in colon tumors, but a detailed report of its upregulation in some specific region of colon epithelia is available where it promotes tumor dissemination.
Nice chart! It could be because 'level' doesn't correlate with 'target' and that a different mechanism is operating. But your test cannot tell you 'why' since it is simply a correlation, not a regression (which indicates cause and effect). You could perhaps try a regression? However, it is usually impossible to determine 'reasons' in experiments, since that is teleological and only God will know why! To get close, you would have to frame your null hypothesis in such a way as to answer that question so that analysis of the appropriate data rules it in or out :)
At the scatter plot, I can see a pattern since the green triangle is ok when we are looking for a correlation: not exists high levels of CDKN1A when we have lower levels of tp53. But at the other side, in the red triangle exists a lack of adjustment in the model of linear correlation since we have, high levels of tp53, but a low level of cdkn1a.
Try to undestand why we can found data in the red triangle, some hipothesis: TP53 is a necessary condition, but not a sufficient condition (other requisite is necessary); the red triangle contains some kind of specific condition (age, sex, ethinic, clinical condition, cancer type as said Mohit Arora ...); existis another agent influencing the process and it is not visible in the scatter.
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue