README.md

In Avery-Lisa/PDscreen: What the Package Does (One Line, Title Case)

PDscreen

What's done from Jan 7 to Feb 28 (Jiayue)

For area = 0: 1. Check treated as 0 left 315 Proteins 2. Check treated as missing value left 317 Proteins

Before applying T-test: There are proteins that only have one group, this make t-test failed. Identify proteins who have two groups and perform statistical test only on those proteins.

For applying T-test: non-transform data vs log-transform data 1. non-transform data: (i) t-test for example protein IGKC and its plot (ii) Perform t-test for the whole proteins and extract their P-value (iii) Use p.adjusted function with flase discovery rate and filter the protein who has adjusted p-value < 0.05, discard the high p-value proteins. (iv) 170 Proteins are statisticallt significant

1. log-transform data: zero intensity peoteins: (i)first time try with log transform and perform t-test, find that around 14.3% zero intensity exist. (ii)the most count for protein who have zero intensity is 20 (20/21=95% zero) and the least count for protein who have zero intensity is 1 (1/21=4.7% zero) (iii)decide to make a plot and barplot for protein who has the middle count 10, I pick IGKC (iv) Check the distribution for both group (IGKC)

t-test: (i) since not a large amount of zero intensity through all the obervations, decide to add a small number 0.001 to our data (ii) do log transformation on our new value data and then can perform t-test (iii) same as non-transform data, we left 129 proteins that are statistically significant

1. Comparision for non-transform and log-transformation: (i) Looking for the Protein who is significant in non-transform data but is not significant in log-transform data (ii) Looking for the Protein who is not significant in non-transform data but is significant in log-transform data (iii) plot for whole data with and without log-transformation (iv) Pick one protein from (i) and plot with and without log-transformation (v) Pick one protein from (ii) and plot with and without log-transformation

Conclusion: need log-transformation, before transformation, the data is right skewed and after log-transformation, the distribution of our data seems to have normal distribution with few outliers. (to do : )

Next Steps

28 Jan To write up a description for (with some individual protein plots): - the effect of applying a log transform to the values - looking at reliability with and without log transform - looking at between group differences with and without log transform

21 Jan 2022 -t-tests / wilcoxon rank sum with and without log transform - function to compute statistical tests for each protein - logistic regression

Going back to the reliability:

7 Jan 2022: Look at the difference between reliability if proteins with Area=0 are treated as 0 vs if they are treated as missing (NA)

Look for candidate biomarkers with the following tests: - t-tests - Wilcoxon test - logistic regression - test of proportion for ANY protein

Control for multiple testing

Which proteins discriminate between groups? Which test(s) are most appropriate?

Programming tasks:

• main.R, encapsulate code as a function for protein screening
• Document the function (using Roxygen https://r-pkgs.org/man.html )

8 October improved the import function

1 October 2021 package initialised

References

A good reference for github and R: https://happygitwithr.com/new-github-first.html

A good reference for writing R packages: https://r-pkgs.org/

Project Goals:

A package to import data from Proteome Discoverer, screen and identify proteins of interest

Step 1. Import & wrangle the data

• import the excel data
• extract protein intensity data (all fields starting with Area)
• identify the protein name from the ascension number
• identify the technical replicates
• recode missing values to zero
• match with the clinical data, identify the technical replicates and recode any missing values to 0
• document all manipulations to a log file.

Package Development Goals:

• write a function, with documentation to import the sampleInformation.csv data. This is just practice function writing so take as an argument the filename and output the number of unique IDS.

Avery-Lisa/PDscreen documentation built on March 18, 2022, 4:28 a.m.