COGG or Correlation Optimization of Genetics and Geodemographics is a novel method where we find the maximum correlation of the principal components obtained from genotype data and the Geodemographic matrix, which consists of Geographical coordinates and external information that influence the genetics of a particular region. In other words, if any external factors such as society, languages, occupations, natural phenomenon has created genetic stratification in the population. In our case, we have chosen data from the Indian subcontinent. India is a land of diversity where the Genetics of the country has been largely influenced by prevalence of caste system and presence of different language families. COGG maximizes the squared correlation of the vector of Principal Component and Geodemographic matrix.
--> File containing Principal Components (we use EIGENSTRAT output)
--> File containing external information along with geographical
coordinates. In our case we use Caste and Language information as
external information
--> value of p, the top principal components to be considered.
This also computes a Canonical Correlation Analysis of CCA, which we name COGG-CCA. COGG is one-sided CCA, we extend both sides in CCA. We now consider top "p" principal components, instead of a vector for the Principal Components. We want to maximize the squared correlation of the Principal Component matrix and the Geodemographic matrix. Canonical Correlation Analysis finds linear combinations of the variables in the Genetics and Geodemographic matrices.
You have to clone/download the codes and run the COGG_Wrapper.m
Edit the COGG_Wrapper.m to include the paths to your files containing the Principal Components in the following
format: SampleID PC1 PC2 PC3 PC4 PC5 .. PCp Population
Where, SampleID is the id of each sample under study and the last column contains the population to which the samples belong. The PC1-PCp, are the number of principal components under consideration.
Also include the link to the file containing external information of geographical coordinates and external information which influence the genetics of those populations. Our methods needs a one-to-one relationship with the external factors and the sample. That is, each sample can belong to only one language group and caste group in our case. Our file is formatted as follows:
Population Name #ofSamples State/Province Region Language Caste Latitude Longitude Dataset
Change line 44 in COGG_Wrapper.m according to the format that you are entering. For example, we have the geography, caste and language information in columns 5,6,7,8 in our format. Change the columns accordingly and note that, the first column should always be the population group name/annotation.
Also you have to edit the value of p and enter the desired number of principal components.
Run it as COGG_Wrapper from the command line or MATLAB IDE.