Here, I'm adding a notebook that looks at the pathways that are captured or represented by models that have training sets from different biological contexts (33-pathway_overlap_biological_contexts.Rmd).
We've trained 5 models with different random seeds for each biological condition. I've deemed pathways to be represented or captured if they are significantly associated with at least one LV in the majority (here, 3 or more) models. Then, I looked at the overlap of these pathways between different conditions.
There's a lot that could be dug into for this analysis for probably a full month! But I've kept it to the two things that jumped out at me:
Training on blood makes it easier to capture very "specialized", mature immune cell gene sets. This makes lots of sense!
Models trained on cancer are "missing" NK cell signatures and it's very unlikely to be due to sample size, because if we look at the random subsampling experiment with approximately the same sample size (n = 8000), NK cell signatures are represented in the majority of models. This result could be due to the lack of NK cell infiltrate/often immunosuppressive nature of tumor microenvironments.
Here, I'm adding a notebook that looks at the pathways that are captured or represented by models that have training sets from different biological contexts (
33-pathway_overlap_biological_contexts.Rmd).We've trained 5 models with different random seeds for each biological condition. I've deemed pathways to be represented or captured if they are significantly associated with at least one LV in the majority (here, 3 or more) models. Then, I looked at the overlap of these pathways between different conditions.
There's a lot that could be dug into for this analysis for probably a full month! But I've kept it to the two things that jumped out at me:
n = 8000), NK cell signatures are represented in the majority of models. This result could be due to the lack of NK cell infiltrate/often immunosuppressive nature of tumor microenvironments.Notebook: 33-pathway_overlap_biological_contexts.nb.zip