search

T-Wisse / MEP_Thomas

This repository serves as the documentation platform for my MEP in TU Delft.
1 stars 0 forks source link

Bayesian classifier purpose #2

Open leilaicruz opened 3 years ago

leilaicruz commented 3 years ago

Here describes why are you proposing this to do , in which context and which literature are you using for this.

T-Wisse commented 3 years ago

My goal is to get an idea of what I need to know to make a prediction on a type of genetic interaction.

A classifier such as the naive bayes is made to predict a class based on attribute information. We could describe the genetic interaction between two genes as any of 4 classes: none (neutral), positive, negative and synthetic lethal. While this does not exactly allow us to predict what would change if we introduce a mutation, it may give us an indication of how well we can predict the type of interaction between 2 genes and which attributes are most important in this prediciton.

We have the genetic interaction dataset to test on. However, it requires attributes of the genetic interaction. These could be protein-protein interaction, number and type of known interactions of the involved genes with other genes, and maybe relative position in common pathways. At the very least I expect the attribute data to be incomplete.

What I wonder if this is a resonable idea and if this has been tried already. I still have to look in to that.

leilaicruz commented 3 years ago

I have done something similar , but first looking at correlations between some attributes and the type of interaction , like for example : number of common interactors between each pairwise interaction type , and common number of go terms , and I found that for some genes you see correlation while if you analyse the whole bulk there is no correlation. You can find the code and the figures , in this repo : https://github.com/leilaicruz/python-modules-for-bioinformatic-analyses/blob/master/output_images/common-go-terms-of-BEM1-based-on-their-type.png

You can focus on different modules and study the different correlations with those attributes and other that you make thing of.

leilaicruz commented 3 years ago

please write here what is the context of use for the bayesian classifier , and then answer why is or why is not a good idea for what you want to do