Use for bayesian integration studies. To optimize integration with use of continuous data measurements of each dataset needs to be grouped in bins based on their value compared to the positive predictive geneset and/or the negative predictive geneset.
To have a optimized bayesian integration. This script automates the binning by optimizing for a high likelihood score in a automated and systematic way.
- It is possible to optimize on highest likelihood overall bins (--score 0)
- Optimize for a total likelihood centered around zero but by having a optimized highest likelihood score per bin (--score 1)
Currently, both methodes only apply a global optimalization, which is in most cases the preferred method of optimalization.
The bin borders can be globally optimized for 2,3,4 or 5 bins. For optimal results we prefer using (--bins 0) option to let the binning tool calculate the most optimal number of bins and borders (--bins 0).
The binning process uses a negative and postive set of which all identifiers should be present in the dataset. If there are missing values present or there is overlap in identifiers between the negative and positive set the program will give a error indicating what the problem is.
reads data file
|
extract genes and values
|
checks double genes
|
get genes from positive and negative
|
check if genes in positive and negative
are present in input data file
|
check if there is no overlap between
positive and negative list
|
sort values while maintaining gene order
|
extract sorted genes to list
|
extract sorted values to list
|
get MIN and MAX values
|
set MIN and MAX borders
|
sweeps all combinations of each bin and remembers the score
|
highest scoring option will be given and plotted
download or pull this repository then:
conda create -n <environmentname> python=2.7
conda activate <environmentname>
conda install -c conda-forge matplotlibpython Binning.py -d <datafile> -n <negativelist> -p <positivelist> -o <outputfilename> [--bins 0/2/3/4] [--type local/global] [--score 0/1]--bins 0: calculates the optimum number of bins with a max of 5 bins, with a high likelihood score (preferred method)
--bins 2/3/4/5 force the code to optimize for this many bins
--score 0: |log2((P/totalP)/(N/totalN))| for optimal likelikhood
--score. 1: log2((P/totalP)/(N/totalN)) for closest to 0 optimalization
P = number of positives in bin
totalP = number of total postives in positivelist
N = number of negatives in bin
totalN = number of total negatives in negativelist
--score 0 (optimal likelihood method)
Number of Gene/identifiers for the positive and negative set
totalP 53
totalN 294
Result: overall score = 8.13212179365 (this is optimal likelihood method so high overall score)
0: (0, 11, 2.734786296106959, 1.0, 101.0)
1: (11, 18, 5.056714390994322, 101.0, 162.0)
2: (18, 347, 0.3406211065510634, 162.0, 5572.0)
Number of positive and negatives in optimized bins
0 pos 6 0 neg 5 1 pos 6 1 neg 1 2 pos 41 2 neg 288
plot with the results:
--score 1 (closest to zero --> 0 <--)
Number of Gene/identifiers for the positive and negative set
totalP 53
totalN 294
Result: overall score = 3.44931450115 (this is result of closest to zero --> 0 <--)
0: (0, 104, 1.1009141949048564, 1.0, 1243.0)
1: (104, 173, 0.6237549837182151, 1243.0, 2441.0)
2: (173, 347, -1.7246453225303384, 2441.0, 5572.0)
Number of positive and negatives in optimized bins
0 pos 29 0 neg 75 1 pos 15 1 neg 54 2 pos 9 2 neg 165
plot with the results:
--type localIs currently not working.
27 October 2020
J.P.M. Coolen and D.R. Garza