(WIP) Implement MEME output subcommand `meme`

[bytewife]

• Fixes #23
• This PR implements several ways of writing MEME files (not all of them are "canon" to the MEME spec, such as the CWM variation). The options are as follows:
  • 'PFM': The position-frequency matrix.
  • 'CWM': The contribution-weight matrix.
  • 'hCWM': The hypothetical contribution-weight matrix; hypothetical contribution scores are the contributions of nucleotides not encoded by the one-hot encoding sequence.
  • 'CWM-PFM': The softmax of the contribution-weight matrix.
  • 'hCWM-PFM': The softmax of the hypothetical contribution-weight matrix.

[bytewife]

@jmschrei It appears that tfmodisco implements subclustering of seqlet clusters. How should we handle these w.r.t. writing motifs to the MEME files?

[bytewife]

Example (modisco_results.CWM.meme):

MEME version 5

ALPHABET= ACGT

Background letter frequencies
A 0.25 C 0.25 G 0.25 T 0.25

MOTIF pattern_0
letter-probability matrix: alength= 4 w= 30 nsites= 1
0.000966 0.001605 0.001199 0.000856
0.000863 0.001477 0.000871 0.001402
0.001114 0.000545 0.001213 0.000839
0.001036 0.001990 0.001227 0.000744
0.002566 0.000326 0.000082 0.000209
-0.000404 0.000548 -0.000335 0.003904
0.000359 0.000382 0.001446 0.000812
0.000310 0.002486 0.000013 0.000915
-0.001650 0.000611 -0.000274 -0.000931
-0.002431 -0.001015 -0.001942 -0.000258
-0.001104 -0.000323 -0.000156 -0.001906
-0.000923 0.000257 0.000545 -0.001862
-0.002585 -0.001050 0.000504 -0.002165
-0.006278 -0.002906 -0.001575 -0.004529
-0.006525 -0.005682 0.000230 -0.003893
0.010032 -0.002446 0.000890 -0.004159
-0.000209 -0.000327 -0.000078 0.061095
-0.000307 0.000133 0.038049 -0.002317
0.046923 -0.000388 -0.000595 -0.000854
-0.006760 -0.001430 0.002318 -0.004597
-0.000112 0.000047 -0.000221 0.058326
-0.000075 0.061763 0.000000 0.000003
0.059826 -0.000072 -0.000006 -0.000103
-0.002650 -0.001453 -0.006538 0.008228
-0.003757 -0.000018 -0.004384 -0.007134
-0.000943 0.003998 -0.000030 -0.000349
-0.003023 0.001721 -0.001096 -0.001723
-0.001865 0.001419 0.000822 -0.001262
-0.002786 -0.001106 -0.000543 0.000054
0.000729 0.000193 0.001019 -0.000507

MOTIF pattern_1
letter-probability matrix: alength= 4 w= 30 nsites= 1
0.001049 0.001619 0.000427 0.000511
-0.000200 0.001497 -0.000001 0.001158
-0.000498 0.000407 0.000163 -0.000873
0.000389 0.000782 0.001914 0.000335
-0.004773 -0.001553 -0.003430 -0.004258
-0.004601 -0.001412 -0.000612 -0.003126
-0.005268 -0.002778 -0.003182 -0.001171
-0.002875 -0.001393 -0.001984 0.028612
0.027435 -0.002664 0.001364 -0.002392
0.065485 -0.000058 -0.000566 -0.000410
-0.003637 -0.002039 -0.003270 0.037624
-0.005678 -0.004789 -0.007113 0.000488
0.006520 -0.002398 -0.000639 -0.002607
0.001827 -0.003912 -0.004117 -0.003664
0.010983 -0.004025 -0.004508 -0.001473
0.007936 -0.003902 -0.003507 -0.001718
-0.001166 0.009161 -0.004876 -0.001414
-0.001216 0.089566 0.000095 0.000006
0.063222 0.000122 0.000036 -0.000758
-0.001174 -0.003339 0.028237 0.000797
0.024157 -0.000106 -0.001504 -0.000790
-0.002020 -0.000069 -0.000733 0.048505
-0.000895 -0.001108 0.059248 -0.003064
-0.001797 -0.002617 0.000276 0.005361
-0.005389 -0.001219 -0.004015 -0.004279
-0.002335 0.000955 -0.000330 -0.001823
-0.002292 0.000080 -0.000658 -0.000764
-0.002291 0.000086 -0.000120 -0.001799
-0.001238 0.002040 0.000819 -0.000516
0.000059 0.000320 -0.000112 -0.000813

MOTIF pattern_2
letter-probability matrix: alength= 4 w= 30 nsites= 1
0.000898 0.002598 0.000942 0.000500
0.001459 0.002100 0.000154 0.000675
0.000583 0.001926 0.001299 0.000154
0.000785 0.000994 0.001907 -0.000990
-0.000691 0.000540 0.000071 -0.001234
0.000047 0.000679 -0.000569 -0.001471
-0.000454 -0.000009 0.000828 -0.000837
-0.001950 0.000279 0.003098 -0.002683
-0.001444 -0.001653 -0.000006 -0.002776
-0.004442 0.002268 -0.002209 -0.005088
0.004961 -0.002721 0.008938 -0.001852
-0.002452 -0.002147 -0.001695 0.011936
0.000027 0.045844 -0.000629 0.000031
0.000057 0.000000 0.000000 0.043926
0.000000 0.000000 0.076969 -0.000147
-0.000493 -0.003102 0.009632 -0.001143
-0.000913 -0.003879 -0.002797 0.001645
-0.000913 -0.002719 -0.003337 0.005109
0.000368 -0.002550 -0.001215 0.007616
0.000958 -0.003784 -0.001084 0.000913
-0.000987 -0.001050 -0.001777 0.005295
0.006877 -0.004218 0.003535 -0.004070
-0.000120 -0.000042 0.000193 0.056402
-0.000554 0.041489 0.000058 0.000563
0.023777 -0.000540 -0.000619 -0.001303
0.000133 -0.001554 0.000041 -0.001371
-0.002286 -0.001094 -0.002247 -0.002995
-0.002717 -0.000872 -0.001313 -0.002747
-0.000462 0.001027 0.000507 0.000987
0.000069 0.000760 0.001237 -0.000024

[bytewife]

@jmschrei Could you also verify that the probability matrices calculated for each datatype is as expected?

[jmschrei]

Let's ignore subpatterns for now. We can add support for that afterwards. Probably it would just be writing the patterns out first and then the subpatterns afterwards.

[jmschrei]

🥳