Computational Genomics 2013 JHU Final Project

Put main notes here: year month day
13.4.12
enhancer data:
sequences around 800bp long
kmer size to use: 20
number of enhancer sequences: 2500
number of negative (not enhancer) sequences: 4000

use java for bloomfilter

-working on javabloomfilter - Kyle 13.4.12
-working on kmers (python) Neighborhoods (hamming distance) - Alessandro 13.4.12
-currently working on hmm tutorial in matlab (profile analysis) - Guannan 13.4.14

-first draft of kmer neighberhood feature extraction. (terrible performance) - Alessandro 13.4.14
-first draft of bloomfilter: Kyle 13.4.17 (runs fast but ~50% accuracy with defaults)
kmer_size | %positive called correctly | % negative called correctly | average difference of +/- kmer calls per read
100     46.00%      52%       5.9293
50      49.75%      50.4%     7.1878
30      54.81%      54.5%     7.8611
20      50.89%      54.5%     8.5579
10      0.16%       99.9%     67.8815
5       0%          100%      0

I think as the kmer size gets small, the number of false positives 
from having a larger negative input file causes everything to be identified
as a negative read (not classified as enhancer).

Either way, the default hash functions end up with only about 50% correctness.