My (2) : 10th Generation Approach - Results

10th Generation Approach - Results

Setting 1: mean = 2.5, disp = 0.05

 

Setting 2: mean = 2.5, disp = 10

Setting 3: mean = 5500, disp = 0.05

Setting 4: mean = 5500, disp = 10

Conclusions:

At sample size of 50:
1. metagenomeSeq performs fairly consistent for all of the settings.

2. GLM - Quasi Poisson model performs very well on the high dispersion (no matters means are), but performs fairly on the low dispersion (no matters means are).

3. GLM - Poisson model performs fairly for all of the settings, except the setting with low mean but high dispersion.

4. GLM - NB model performs very well on the high dispersion (no matters means are), but performs  fairly well on the low dispersion (no matters means are).

5. Elastic net: the way that elastic net deal with the data is similar to NB model. That is it performs very well on the high dispersion (no matters means are), but performs  fairly well on the low dispersion (no matters means are).

*** Applying elastic net help to improve the performance of GLM - Poisson model with the setting of low dispersion, but it does not help to improve the performance of GLM - NB model ***