I'm reminded of the last part of 'TLA' form the Jargon File (I had a hard copy back in college that I read cover to cover).
http://www.catb.org/jargon/html/T/TLA.html
...
The self-effacing phrase “TDM TLA” (Too Damn Many...) is often used to bemoan the plethora of TLAs in use. In 1989, a random of the journalistic persuasion asked hacker Paul Boutin “What do you think will be the biggest problem in computing in the 90s?” Paul's straight-faced response: “There are only 17,000 three-letter acronyms.” (To be exact, there are 26^3 = 17,576.) There is probably some karmic justice in the fact that Paul Boutin subsequently became a journalist.
Now I want to use the dictionary file to figure the actual probability of a letter appearing in a TLA. It's not nearly 1/26.
There's likely a good bit of analysis that could be done on TLAs. Consider TLA itself is {Adjective : Count} {Noun} {Noun}. Meanwhile, DUI is {Gerund} {Preposition} {Noun} with the stop word 'the' removed.
It might be interesting to take a sample of TLAs used and look what words can be used in those spots. If the third position is 90% likely to be a noun, that could change the distribution... guessing not in a significant way itself but it could be interesting to see.
This is the best work I know on the topic (admittedly having done no literature review): https://gwern.net/tla