+ added splicesite scores to training
[qpalma.git] / qpalma / computeSpliceWeights.py
1 #!/usr/bin/env python
2 # -*- coding: utf-8 -*-
3
4 from numpy import inf
5 from numpy.matlib import zeros
6 import pdb
7
8 # 3 * 30 supporting points: x_1 ... x_30 => 3 * 30 parameters (1 .. 90): y_1 ... y_30
9 # piecewise linear function:
10 # take score from SVM vektor (don_supp: case 1, acc_supp: case 2) and compute length of intron: case 3
11 # these are our values x
12 #
13 # | y_1 if x <= x_1
14 # |
15 # | x_i+1 - x x - x_i
16 # f(x) = | y_i * ----------- + y_i+1 * ----------- if x_i <= x <= x_i+1
17 # | x_i+1 - x_i x_i+1 - x_i
18 # |
19 # | y_30 if x_n <= x
20 #
21 # y_i and y_i+1 parameters, so the fractions are saved in the weight vectors!
22
23 def calculateWeights(plf, scores):
24 currentWeight = zeros((30,1))
25
26 for k in range(len(scores)):
27 value = scores[k]
28 Lower = len([elem for elem in plf.limits if elem <= value])
29
30 if Lower == 0:
31 currentWeight[0] += 1
32 elif Lower == len(plf.limits):
33 currentWeight[-1] += 1
34 else:
35 # because we count from 0 in python
36 Lower -= 1
37 Upper = Lower+1 ; # x-werte bleiben fest
38 #print value,Lower,Upper
39 weightup = 1.0*(value - plf.limits[Lower]) / (plf.limits[Upper] - plf.limits[Lower])
40 weightlow = 1.0*(plf.limits[Upper] - value) / (plf.limits[Upper] - plf.limits[Lower])
41 currentWeight[Upper] = currentWeight[Upper] + weightup
42 currentWeight[Lower] = currentWeight[Lower] + weightlow
43
44 #print plf.limits[Lower],plf.limits[Upper]
45 #print weightup,weightlow,currentWeight[Upper],currentWeight[Lower]
46
47 return currentWeight
48
49 def computeSpliceWeights(d, a, h, SpliceAlign, don_supp, acc_supp,dec=False):
50 ####################################################################################
51 # 1. Donor: In don_supp stehen Werte der SVM., in SpliceAlign die Einsen
52 ####################################################################################
53
54 # Picke die Positionen raus, an denen eine Donorstelle ist
55 try:
56 if dec:
57 DonorScores = [elem for pos,elem in enumerate(don_supp) if pos > 0 and SpliceAlign[pos] == 1]
58 else:
59 DonorScores = [elem for pos,elem in enumerate(don_supp) if pos > 0 and SpliceAlign[pos-1] == 1]
60 assert not ( -inf in DonorScores )
61 except:
62 pdb.set_trace()
63
64 #print 'donor'
65 weightDon = calculateWeights(d,DonorScores)
66
67 ####################################################################################
68 # 2. Acceptor: In acc_supp stehen Werte der SVM., in SpliceAlign die Einsen
69 ####################################################################################
70
71 #Den Vektor Acceptorstellen durchgehen und die Gewichtsvektoren belasten:
72 try:
73 if dec:
74 AcceptorScores = [elem for pos,elem in enumerate(acc_supp) if pos > 0 and SpliceAlign[pos-1] == 2]
75 else:
76 AcceptorScores = [elem for pos,elem in enumerate(acc_supp) if pos > 0 and SpliceAlign[pos] == 2]
77 assert not ( -inf in AcceptorScores )
78 except:
79 pdb.set_trace()
80
81 #print 'acceptor'
82 weightAcc = calculateWeights(a,AcceptorScores)
83
84 ####################################################################################
85 # 3. Intron length: SpliceAlign: Gaps zaehlen und auf Gapgewichte addieren
86 ####################################################################################
87
88 intron_starts = []
89 intron_ends = []
90 for pos,elem in enumerate(SpliceAlign):
91 if elem == 1:
92 intron_starts.append(pos)
93
94 if elem == 2:
95 intron_ends.append(pos)
96
97 assert len(intron_starts) == len(intron_ends)
98
99 for i in range(len(intron_starts)):
100 assert intron_starts[i] < intron_ends[i]
101
102 values = [0.0]*len(intron_starts)
103 for pos in range(len(intron_starts)):
104 values[pos] = intron_ends[pos] - intron_starts[pos] + 1
105
106 weightIntron = calculateWeights(h,values)
107
108 return weightDon, weightAcc, weightIntron