TODO

SELEX.py

1) Build a rnds and sequences object
   - should have a code sequences method
   - should validate sequence lengths, and store initial condition information
2) clean up motif object to store ddg infromation, etc.
   - write a __str__ method
3) Get rid of the stupid matrix multiplication, and write a proper cude kernel

Questions:
How do in-vitro compare to 'known' motifs
    -What's the correlation with ATAC-seq in the upper rank list
    -Do multiple motifs improve our prediction accuracy
    -Write a script to find the best motif
How do motifs from different factors do
    -Anshul
Can we predict the chemical affinity *without* using ChIP-seq?

TODO:
Find the best base gibbs free energy for each motif
     - Maximize marginal rank correlation with ChIP-seq  
Optimize over the best motif
Use DNASE instead of ATAC-seq 
Include chromatin marks 
do a better job with fragment elngth distribution
   - use mean fragment length from cross correlation analysis