Clover icon

Coverage Report

  1. Project Clover database Wed Nov 13 2024 18:27:33 GMT
  2. Package jalview.analysis

File StructureFrequency.java

 

Coverage histogram

../../img/srcFileCovDistChart6.png
36% of files have more coverage

Code metrics

90
173
4
1
470
319
58
0.34
43.25
4
14.5

Classes

Class Line # Actions
StructureFrequency 41 173 58
0.580524358.1%
 

Contributing tests

This file is covered by 1 test. .

Source view

1    /*
2    * Jalview - A Sequence Alignment Editor and Viewer ($$Version-Rel$$)
3    * Copyright (C) $$Year-Rel$$ The Jalview Authors
4    *
5    * This file is part of Jalview.
6    *
7    * Jalview is free software: you can redistribute it and/or
8    * modify it under the terms of the GNU General Public License
9    * as published by the Free Software Foundation, either version 3
10    * of the License, or (at your option) any later version.
11    *
12    * Jalview is distributed in the hope that it will be useful, but
13    * WITHOUT ANY WARRANTY; without even the implied warranty
14    * of MERCHANTABILITY or FITNESS FOR A PARTICULAR
15    * PURPOSE. See the GNU General Public License for more details.
16    *
17    * You should have received a copy of the GNU General Public License
18    * along with Jalview. If not, see <http://www.gnu.org/licenses/>.
19    * The Jalview Authors are detailed in the 'AUTHORS' file.
20    */
21    package jalview.analysis;
22   
23    import jalview.datamodel.AlignmentAnnotation;
24    import jalview.datamodel.Annotation;
25    import jalview.datamodel.SequenceFeature;
26    import jalview.datamodel.SequenceI;
27    import jalview.util.Comparison;
28    import jalview.util.Format;
29   
30    import java.util.Hashtable;
31   
32    /**
33    * Takes in a vector or array of sequences and column start and column end and
34    * returns a new Hashtable[] of size maxSeqLength, if Hashtable not supplied.
35    * This class is used extensively in calculating alignment colourschemes that
36    * depend on the amount of conservation in each alignment column.
37    *
38    * @author $author$
39    * @version $Revision$
40    */
 
41    public class StructureFrequency
42    {
43    public static final int STRUCTURE_PROFILE_LENGTH = 74;
44   
45    // No need to store 1000s of strings which are not
46    // visible to the user.
47    public static final String MAXCOUNT = "C";
48   
49    public static final String MAXRESIDUE = "R";
50   
51    public static final String PID_GAPS = "G";
52   
53    public static final String PID_NOGAPS = "N";
54   
55    public static final String PROFILE = "P";
56   
57    public static final String PAIRPROFILE = "B";
58   
59    /**
60    * Returns the 3' position of a base pair
61    *
62    * @param pairs
63    * Secondary structure annotation
64    * @param indice
65    * 5' position of a base pair
66    * @return 3' position of a base pair
67    */
 
68  90 toggle public static int findPair(SequenceFeature[] pairs, int indice)
69    {
70   
71  855 for (int i = 0; i < pairs.length; i++)
72    {
73  855 if (pairs[i].getBegin() == indice)
74   
75    {
76   
77  90 return pairs[i].getEnd();
78   
79    }
80    }
81  0 return -1;
82    }
83   
84    /**
85    * Method to calculate a 'base pair consensus row', very similar to nucleotide
86    * consensus but takes into account a given structure
87    *
88    * @param sequences
89    * @param start
90    * @param end
91    * @param result
92    * @param profile
93    * @param rnaStruc
94    */
 
95  5 toggle public static final void calculate(SequenceI[] sequences, int start,
96    int end, Hashtable<String, Object>[] result, boolean profile,
97    AlignmentAnnotation rnaStruc)
98    {
99   
100  5 Hashtable<String, Object> residueHash;
101  5 String maxResidue;
102  5 char[] struc = rnaStruc.getRNAStruc().toCharArray();
103   
104  5 SequenceFeature[] rna = rnaStruc._rnasecstr;
105  5 char c, s, cEnd;
106  5 int bpEnd = -1;
107  5 int jSize = sequences.length;
108  5 int[] values;
109  5 int[][] pairs;
110  5 float percentage;
111   
112  470 for (int i = start; i < end; i++) // foreach column
113    {
114  465 int canonicalOrWobblePairCount = 0, canonical = 0;
115  465 int otherPairCount = 0;
116  465 int nongap = 0;
117  465 maxResidue = "-";
118  464 values = new int[255];
119  465 pairs = new int[255][255];
120  465 bpEnd = -1;
121  465 if (i < struc.length)
122    {
123  465 s = struc[i];
124    }
125    else
126    {
127  0 s = '-';
128    }
129  465 if (s == '.' || s == ' ')
130    {
131  285 s = '-';
132    }
133   
134  465 if (!Rna.isOpeningParenthesis(s))
135    {
136  375 if (s == '-')
137    {
138  285 values['-']++;
139    }
140    }
141    else
142    {
143  90 bpEnd = findPair(rna, i);
144   
145  90 if (bpEnd > -1)
146    {
147  5579 for (int j = 0; j < jSize; j++) // foreach row
148    {
149  5490 if (sequences[j] == null)
150    {
151  0 jalview.bin.Console.errPrintln(
152    "WARNING: Consensus skipping null sequence - possible race condition.");
153  0 continue;
154    }
155   
156  5490 c = sequences[j].getCharAt(i);
157  5490 cEnd = sequences[j].getCharAt(bpEnd);
158   
159  5489 if (Comparison.isGap(c) || Comparison.isGap(cEnd))
160    {
161  10 values['-']++;
162  10 continue;
163    }
164  5480 nongap++;
165    /*
166    * ensure upper-case for counting purposes
167    */
168  5480 if ('a' <= c && 'z' >= c)
169    {
170  0 c += 'A' - 'a';
171    }
172  5480 if ('a' <= cEnd && 'z' >= cEnd)
173    {
174  0 cEnd += 'A' - 'a';
175    }
176  5480 if (Rna.isCanonicalOrWobblePair(c, cEnd))
177    {
178  4865 canonicalOrWobblePairCount++;
179  4865 if (Rna.isCanonicalPair(c, cEnd))
180    {
181  4315 canonical++;
182    }
183    }
184    else
185    {
186  615 otherPairCount++;
187    }
188  5480 pairs[c][cEnd]++;
189    }
190    }
191    }
192   
193  465 residueHash = new Hashtable<>();
194  465 if (profile)
195    {
196    // TODO 1-dim array with jsize in [0], nongapped in [1]; or Pojo
197  465 residueHash.put(PROFILE,
198    new int[][]
199    { values, new int[] { jSize, (jSize - values['-']) } });
200   
201  465 residueHash.put(PAIRPROFILE, pairs);
202    }
203  465 values['('] = canonicalOrWobblePairCount;
204  465 values['['] = canonical;
205  465 values['{'] = otherPairCount;
206    /*
207    * the count is the number of valid pairs (as a percentage, determines
208    * the relative size of the profile logo)
209    */
210  465 int count = canonicalOrWobblePairCount;
211   
212    /*
213    * display '(' if most pairs are canonical, or as
214    * '[' if there are more wobble pairs.
215    */
216  465 if (canonicalOrWobblePairCount > 0 || otherPairCount > 0)
217    {
218  90 if (canonicalOrWobblePairCount >= otherPairCount)
219    {
220  90 maxResidue = (canonicalOrWobblePairCount - canonical) < canonical
221    ? "("
222    : "[";
223    }
224    else
225    {
226  0 maxResidue = "{";
227    }
228    }
229  465 residueHash.put(MAXCOUNT, Integer.valueOf(count));
230  465 residueHash.put(MAXRESIDUE, maxResidue);
231   
232  465 percentage = ((float) count * 100) / jSize;
233  465 residueHash.put(PID_GAPS, Float.valueOf(percentage));
234   
235  465 percentage = ((float) count * 100) / nongap;
236  465 residueHash.put(PID_NOGAPS, Float.valueOf(percentage));
237   
238  465 if (result[i] == null)
239    {
240  375 result[i] = residueHash;
241    }
242  465 if (bpEnd > 0)
243    {
244  90 values[')'] = values['('];
245  90 values[']'] = values['['];
246  90 values['}'] = values['{'];
247  90 values['('] = 0;
248  90 values['['] = 0;
249  90 values['{'] = 0;
250  90 maxResidue = maxResidue.equals("(") ? ")"
251  0 : maxResidue.equals("[") ? "]" : "}";
252   
253  90 residueHash = new Hashtable<>();
254  90 if (profile)
255    {
256  90 residueHash.put(PROFILE,
257    new int[][]
258    { values, new int[] { jSize, (jSize - values['-']) } });
259   
260  90 residueHash.put(PAIRPROFILE, pairs);
261    }
262   
263  90 residueHash.put(MAXCOUNT, Integer.valueOf(count));
264  90 residueHash.put(MAXRESIDUE, maxResidue);
265   
266  90 percentage = ((float) count * 100) / jSize;
267  90 residueHash.put(PID_GAPS, Float.valueOf(percentage));
268   
269  90 percentage = ((float) count * 100) / nongap;
270  90 residueHash.put(PID_NOGAPS, Float.valueOf(percentage));
271   
272  90 result[bpEnd] = residueHash;
273    }
274    }
275    }
276   
277    /**
278    * Compute all or part of the annotation row from the given consensus
279    * hashtable
280    *
281    * @param consensus
282    * - pre-allocated annotation row
283    * @param hconsensus
284    * @param iStart
285    * @param width
286    * @param ignoreGapsInConsensusCalculation
287    * @param includeAllConsSymbols
288    */
 
289  5 toggle public static void completeConsensus(AlignmentAnnotation consensus,
290    Hashtable<String, Object>[] hconsensus, int iStart, int width,
291    boolean ignoreGapsInConsensusCalculation,
292    boolean includeAllConsSymbols, long nseq)
293    {
294  5 float tval, value;
295  5 if (consensus == null || consensus.annotations == null
296    || consensus.annotations.length < width)
297    {
298    // called with a bad alignment annotation row - wait for it to be
299    // initialised properly
300  0 return;
301    }
302  5 String fmtstr = "%3.1f";
303  5 int precision = 2;
304  5 while (nseq > 100)
305    {
306  0 precision++;
307  0 nseq /= 10;
308    }
309  5 if (precision > 2)
310    {
311  0 fmtstr = "%" + (2 + precision) + "." + precision + "f";
312    }
313  5 Format fmt = new Format(fmtstr);
314   
315  470 for (int i = iStart; i < width; i++)
316    {
317  465 Hashtable<String, Object> hci;
318  ? if (i >= hconsensus.length || ((hci = hconsensus[i]) == null))
319    {
320    // happens if sequences calculated over were shorter than alignment
321    // width
322  0 consensus.annotations[i] = null;
323  0 continue;
324    }
325  465 value = 0;
326  465 Float fv;
327  465 if (ignoreGapsInConsensusCalculation)
328    {
329  0 fv = (Float) hci.get(StructureFrequency.PID_NOGAPS);
330    }
331    else
332    {
333  465 fv = (Float) hci.get(StructureFrequency.PID_GAPS);
334    }
335  465 if (fv == null)
336    {
337  0 consensus.annotations[i] = null;
338    // data has changed below us .. give up and
339  0 continue;
340    }
341  465 value = fv.floatValue();
342  465 String maxRes = hci.get(StructureFrequency.MAXRESIDUE).toString();
343  465 String mouseOver = hci.get(StructureFrequency.MAXRESIDUE) + " ";
344  465 if (maxRes.length() > 1)
345    {
346  0 mouseOver = "[" + maxRes + "] ";
347  0 maxRes = "+";
348    }
349  465 int[][] profile = (int[][]) hci.get(StructureFrequency.PROFILE);
350  465 int[][] pairs = (int[][]) hci.get(StructureFrequency.PAIRPROFILE);
351   
352  465 if (pairs != null && includeAllConsSymbols) // Just responsible for the
353    // tooltip
354    // TODO Update tooltips for Structure row
355    {
356  0 mouseOver = "";
357   
358    /*
359    * TODO It's not sure what is the purpose of the alphabet and wheter it
360    * is useful for structure?
361    *
362    * if (alphabet != null) { for (int c = 0; c < alphabet.length; c++) {
363    * tval = ((float) profile[0][alphabet[c]]) 100f / (float)
364    * profile[1][ignoreGapsInConsensusCalculation ? 1 : 0]; mouseOver +=
365    * ((c == 0) ? "" : "; ") + alphabet[c] + " " + ((int) tval) + "%"; } }
366    * else {
367    */
368  0 int[][] ca = new int[625][];
369  0 float[] vl = new float[625];
370  0 int x = 0;
371  0 for (int c = 65; c < 90; c++)
372    {
373  0 for (int d = 65; d < 90; d++)
374    {
375  0 ca[x] = new int[] { c, d };
376  0 vl[x] = pairs[c][d];
377  0 x++;
378    }
379    }
380  0 jalview.util.QuickSort.sort(vl, ca);
381  0 int p = 0;
382   
383    /*
384    * profile[1] is {total, ungappedTotal}
385    */
386  0 final int divisor = profile[1][ignoreGapsInConsensusCalculation ? 1
387    : 0];
388  0 for (int c = 624; c > 0; c--)
389    {
390  0 if (vl[c] > 0)
391    {
392  0 tval = (vl[c] * 100f / divisor);
393  0 mouseOver += ((p == 0) ? "" : "; ") + (char) ca[c][0]
394    + (char) ca[c][1] + " " + fmt.form(tval) + "%";
395  0 p++;
396   
397    }
398    }
399   
400    // }
401    }
402    else
403    {
404  465 mouseOver += (fmt.form(value) + "%");
405    }
406  465 consensus.annotations[i] = new Annotation(maxRes, mouseOver, ' ',
407    value);
408    }
409    }
410   
411    /**
412    * get the sorted base-pair profile for the given position of the consensus
413    *
414    * @param hconsensus
415    * @return profile of the given column
416    */
 
417  0 toggle public static int[] extractProfile(Hashtable<String, Object> hconsensus,
418    boolean ignoreGapsInConsensusCalculation)
419    {
420  0 int[] rtnval = new int[STRUCTURE_PROFILE_LENGTH]; // 2*(5*5)+2
421  0 int[][] profile = (int[][]) hconsensus.get(StructureFrequency.PROFILE);
422  0 int[][] pairs = (int[][]) hconsensus
423    .get(StructureFrequency.PAIRPROFILE);
424   
425  0 if (profile == null)
426    {
427  0 return null;
428    }
429   
430    // TODO fix the object length, also do it in completeConsensus
431    // Object[] ca = new Object[625];
432  0 int[][] ca = new int[625][];
433  0 float[] vl = new float[625];
434  0 int x = 0;
435  0 for (int c = 65; c < 90; c++)
436    {
437  0 for (int d = 65; d < 90; d++)
438    {
439  0 ca[x] = new int[] { c, d };
440  0 vl[x] = pairs[c][d];
441  0 x++;
442    }
443    }
444  0 jalview.util.QuickSort.sort(vl, ca);
445   
446  0 int valuesCount = 0;
447  0 rtnval[1] = 0;
448  0 int offset = 2;
449  0 final int divisor = profile[1][ignoreGapsInConsensusCalculation ? 1
450    : 0];
451  0 for (int c = 624; c > 0; c--)
452    {
453  0 if (vl[c] > 0)
454    {
455  0 rtnval[offset++] = ca[c][0];
456  0 rtnval[offset++] = ca[c][1];
457  0 rtnval[offset] = (int) (vl[c] * 100f / divisor);
458  0 rtnval[1] += rtnval[offset++];
459  0 valuesCount++;
460    }
461    }
462  0 rtnval[0] = valuesCount;
463   
464    // insert profile type code in position 0
465  0 int[] result = new int[rtnval.length + 1];
466  0 result[0] = AlignmentAnnotation.STRUCTURE_PROFILE;
467  0 System.arraycopy(rtnval, 0, result, 1, rtnval.length);
468  0 return result;
469    }
470    }