To piggyback, if you are interested in personalized cancer vaccines, we're also hiring at WashU for a senior scientist position that will be working at least part time on immunotherapy and personalized cancer vaccines. If you're an interdisciplinary person, who knows cancer and can code, drop me a line (c.a.miller at wustl.edu)
For a very long long time, Pathways are the word used by biologists for defined signalling cascades inside cells, to used the same word for a clustered gene set is quite confusing.
Your approach is very data driven and as it is you are ignoring a vast sea of literature out there that investigates the relation between signalling pathways and target mRNAs by doing research into (among others) transcription complexes directly (i.e. using CHIP [0]) (like they do i.e. here: [1]). I get that for computer scientists this is perhaps the fastest way to results but I feel such research can be done in a smarter way, by teaming up with biologists. Don't you agree?
[0] https://en.wikipedia.org/wiki/Chromatin_immunoprecipitation
A promising path would be to self-teach as much as possible and then apply for an MS program in bioinformatics or something similar.
Apart from the usual advice for teaching yourself programming, you can practice bioinformatic-specific skills at the wonderful Rosalind Project [0].
You already know some biology, so you can probably say more about that side of the equation.
If you are an experienced machine learning engineer and would like to help improve the quality of our vaccines, please send me an email at hammer at hammerlab dot org. We are located in NYC.