rajewsky-lab / polyA

0 stars 1 forks source link

switched simulation from fixed fragment size to distribution #63

Closed mschilli87 closed 8 years ago

mschilli87 commented 8 years ago

should work as is, but changing f_size_sim = np.array([400]) to f_size_sim = np.array([399,400]) and f_prob_sim = np.array([1]) to f_prob_sim = np.array([.1,.9]) results in a shift of est_pAlen from the expected 42 to 43.

codecov-io commented 8 years ago

Current coverage is 73.06%

Merging #63 into simulation will increase coverage by 0.98%

@@           simulation        #63   diff @@
============================================
  Files               3          3          
  Lines             308        323    +15   
  Methods             0          0          
  Messages            0          0          
  Branches            0          0          
============================================
+ Hits              222        236    +14   
- Misses             86         87     +1   
  Partials            0          0          

Powered by Codecov. Last updated by b5954bc...7f37470

nukappa commented 8 years ago

i checked this also. played with different reads and/or bioanalyzer "profiles" and model behaves as it should... iguess you'll push more things here so i leave it for now?

mschilli87 commented 8 years ago

Yes, you can leave it around for now. Do you have any clue why the shift to 43 happens?


edit: Opened #66 to further discuss that issue.

mschilli87 commented 8 years ago

I merged in current master to be able to work on this more efficiently thanks to https://github.com/rajewsky-lab/polyA/commit/0e362b1e04be027d48df75f234c1110a3ea9ce81.

mschilli87 commented 8 years ago

Simulating random annealing of the oligo-d(T) to the poly(A) breaks the 42 expected value for the fixed 400nt insert size case (I've seen a single 41 in a local run) breaking the corresponding unit test but it seems to greatly improve the 90/10-95/100nt cases (a few 41s and 43s amongs many 42s) as we suspected. I'll continue with real profiles and so on this branch but I guess we might have to loosen the unit test criterion a bit eventually. Any suggestions? Person correlation (p-value)?

nukappa commented 8 years ago

very nice! yes, please continue with real profiles. we'll think about the unit test later..

codecov-io commented 8 years ago

Current coverage is 73.17%

Merging #63 into simulation will increase coverage by 1.09%

@@           simulation        #63   diff @@
============================================
  Files               3          3          
  Lines             308        328    +20   
  Methods             0          0          
  Messages            0          0          
  Branches            0          0          
============================================
+ Hits              222        240    +18   
- Misses             86         88     +2   
  Partials            0          0          

Powered by Codecov. Last updated by b5954bc...38cab50