2 * Jalview - A Sequence Alignment Editor and Viewer ($$Version-Rel$$)
3 * Copyright (C) $$Year-Rel$$ The Jalview Authors
5 * This file is part of Jalview.
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.
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.
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.
21 package jalview.analysis;
23 import jalview.datamodel.AlignedCodonFrame;
24 import jalview.datamodel.AlignmentAnnotation;
25 import jalview.datamodel.AlignmentI;
26 import jalview.datamodel.Annotation;
27 import jalview.datamodel.HiddenMarkovModel;
28 import jalview.datamodel.Profile;
29 import jalview.datamodel.ProfileI;
30 import jalview.datamodel.Profiles;
31 import jalview.datamodel.ProfilesI;
32 import jalview.datamodel.ResidueCount;
33 import jalview.datamodel.ResidueCount.SymbolCounts;
34 import jalview.datamodel.SequenceI;
35 import jalview.ext.android.SparseIntArray;
36 import jalview.schemes.ResidueProperties;
37 import jalview.util.Comparison;
38 import jalview.util.Format;
39 import jalview.util.MappingUtils;
40 import jalview.util.QuickSort;
42 import java.awt.Color;
43 import java.util.Arrays;
44 import java.util.Hashtable;
45 import java.util.List;
48 * Takes in a vector or array of sequences and column start and column end and
49 * returns a new Hashtable[] of size maxSeqLength, if Hashtable not supplied.
50 * This class is used extensively in calculating alignment colourschemes that
51 * depend on the amount of conservation in each alignment column.
54 public class AAFrequency
56 private static final double LOG2 = Math.log(2);
58 public static final String PROFILE = "P";
60 public static final ProfilesI calculate(List<SequenceI> list, int start,
63 return calculate(list, start, end, false);
66 public static final ProfilesI calculate(List<SequenceI> sequences,
67 int start, int end, boolean profile)
69 SequenceI[] seqs = new SequenceI[sequences.size()];
71 synchronized (sequences)
73 for (int i = 0; i < sequences.size(); i++)
75 seqs[i] = sequences.get(i);
76 int length = seqs[i].getLength();
88 ProfilesI reply = calculate(seqs, width, start, end, profile);
94 * Calculate the consensus symbol(s) for each column in the given range.
98 * the full width of the alignment
100 * start column (inclusive, base zero)
102 * end column (exclusive)
103 * @param saveFullProfile
104 * if true, store all symbol counts
106 public static final ProfilesI calculate(final SequenceI[] sequences,
107 int width, int start, int end, boolean saveFullProfile)
109 // long now = System.currentTimeMillis();
110 int seqCount = sequences.length;
111 boolean nucleotide = false;
112 int nucleotideCount = 0;
113 int peptideCount = 0;
115 ProfileI[] result = new ProfileI[width];
117 for (int column = start; column < end; column++)
120 * Apply a heuristic to detect nucleotide data (which can
121 * be counted in more compact arrays); here we test for
122 * more than 90% nucleotide; recheck every 10 columns in case
123 * of misleading data e.g. highly conserved Alanine in peptide!
124 * Mistakenly guessing nucleotide has a small performance cost,
125 * as it will result in counting in sparse arrays.
126 * Mistakenly guessing peptide has a small space cost,
127 * as it will use a larger than necessary array to hold counts.
129 if (nucleotideCount > 100 && column % 10 == 0)
131 nucleotide = (9 * peptideCount < nucleotideCount);
133 ResidueCount residueCounts = new ResidueCount(nucleotide);
135 for (int row = 0; row < seqCount; row++)
137 if (sequences[row] == null)
140 "WARNING: Consensus skipping null sequence - possible race condition.");
143 if (sequences[row].getLength() > column)
145 char c = sequences[row].getCharAt(column);
146 residueCounts.add(c);
147 if (Comparison.isNucleotide(c))
151 else if (!Comparison.isGap(c))
159 * count a gap if the sequence doesn't reach this column
161 residueCounts.addGap();
165 int maxCount = residueCounts.getModalCount();
166 String maxResidue = residueCounts.getResiduesForCount(maxCount);
167 int gapCount = residueCounts.getGapCount();
168 ProfileI profile = new Profile(seqCount, gapCount, maxCount,
173 profile.setCounts(residueCounts);
176 result[column] = profile;
178 return new Profiles(result);
179 // long elapsed = System.currentTimeMillis() - now;
180 // System.out.println(elapsed);
184 * Returns the full set of profiles for a hidden Markov model. The underlying
185 * data is the raw probabilities of a residue being emitted at each node,
186 * however the profiles returned by this function contain the percentage
187 * chance of a residue emission.
191 * The width of the Profile array (Profiles) to be returned.
193 * The alignment column on which the first profile is based.
195 * The alignment column on which the last profile is based.
196 * @param saveFullProfile
197 * Flag for saving the counts for each profile
198 * @param removeBelowBackground
199 * Flag for removing any characters with a match emission probability
200 * less than its background frequency
203 public static ProfilesI calculateHMMProfiles(final HiddenMarkovModel hmm,
204 int width, int start, int end, boolean saveFullProfile,
205 boolean removeBelowBackground, boolean infoLetterHeight)
207 ProfileI[] result = new ProfileI[width];
208 char[] symbols = hmm.getSymbols().toCharArray();
209 int symbolCount = symbols.length;
210 for (int column = start; column < end; column++)
212 ResidueCount counts = new ResidueCount();
213 for (char symbol : symbols)
215 int value = getAnalogueCount(hmm, column, symbol,
216 removeBelowBackground, infoLetterHeight);
217 counts.put(symbol, value);
219 int maxCount = counts.getModalCount();
220 String maxResidue = counts.getResiduesForCount(maxCount);
221 int gapCount = counts.getGapCount();
222 ProfileI profile = new Profile(symbolCount, gapCount, maxCount,
227 profile.setCounts(counts);
230 result[column] = profile;
232 return new Profiles(result);
236 * Make an estimate of the profile size we are going to compute i.e. how many
237 * different characters may be present in it. Overestimating has a cost of
238 * using more memory than necessary. Underestimating has a cost of needing to
239 * extend the SparseIntArray holding the profile counts.
241 * @param profileSizes
242 * counts of sizes of profiles so far encountered
245 static int estimateProfileSize(SparseIntArray profileSizes)
247 if (profileSizes.size() == 0)
253 * could do a statistical heuristic here e.g. 75%ile
254 * for now just return the largest value
256 return profileSizes.keyAt(profileSizes.size() - 1);
260 * Derive the consensus annotations to be added to the alignment for display.
261 * This does not recompute the raw data, but may be called on a change in
262 * display options, such as 'ignore gaps', which may in turn result in a
263 * change in the derived values.
266 * the annotation row to add annotations to
268 * the source consensus data
270 * start column (inclusive)
272 * end column (exclusive)
274 * if true, normalise residue percentages ignoring gaps
275 * @param showSequenceLogo
276 * if true include all consensus symbols, else just show modal
279 * number of sequences
281 public static void completeConsensus(AlignmentAnnotation consensus,
282 ProfilesI profiles, int startCol, int endCol, boolean ignoreGaps,
283 boolean showSequenceLogo, long nseq)
285 // long now = System.currentTimeMillis();
286 if (consensus == null || consensus.annotations == null
287 || consensus.annotations.length < endCol)
290 * called with a bad alignment annotation row
291 * wait for it to be initialised properly
296 for (int i = startCol; i < endCol; i++)
298 ProfileI profile = profiles.get(i);
302 * happens if sequences calculated over were
303 * shorter than alignment width
305 consensus.annotations[i] = null;
309 final int dp = getPercentageDp(nseq);
311 float value = profile.getPercentageIdentity(ignoreGaps);
313 String description = getTooltip(profile, value, showSequenceLogo,
316 String modalResidue = profile.getModalResidue();
317 if ("".equals(modalResidue))
321 else if (modalResidue.length() > 1)
325 consensus.annotations[i] = new Annotation(modalResidue, description,
328 // long elapsed = System.currentTimeMillis() - now;
329 // System.out.println(-elapsed);
333 * Derive the information annotations to be added to the alignment for
334 * display. This does not recompute the raw data, but may be called on a
335 * change in display options, such as 'ignore below background frequency',
336 * which may in turn result in a change in the derived values.
339 * the annotation row to add annotations to
341 * the source information data
343 * start column (inclusive)
345 * end column (exclusive)
347 * if true, normalise residue percentages
348 * @param showSequenceLogo
349 * if true include all information symbols, else just show modal
352 * number of sequences
354 public static float completeInformation(AlignmentAnnotation information,
355 ProfilesI profiles, int startCol, int endCol, long nseq,
358 // long now = System.currentTimeMillis();
359 if (information == null || information.annotations == null
360 || information.annotations.length < endCol)
363 * called with a bad alignment annotation row
364 * wait for it to be initialised properly
371 for (int i = startCol; i < endCol; i++)
373 ProfileI profile = profiles.get(i);
377 * happens if sequences calculated over were
378 * shorter than alignment width
380 information.annotations[i] = null;
384 SequenceI hmmSeq = information.sequenceRef;
386 HiddenMarkovModel hmm = hmmSeq.getHMM();
388 float value = hmm.getInformationContent(i);
395 String description = value + " bits";
396 information.annotations[i] = new Annotation(
397 Character.toString(Character
398 .toUpperCase(hmm.getConsensusAtAlignColumn(i))),
399 description, ' ', value);
401 if (max > currentMax)
403 information.graphMax = max;
408 information.graphMax = currentMax;
414 * Derive the gap count annotation row.
417 * the annotation row to add annotations to
419 * the source consensus data
421 * start column (inclusive)
423 * end column (exclusive)
425 public static void completeGapAnnot(AlignmentAnnotation gaprow,
426 ProfilesI profiles, int startCol, int endCol, long nseq)
428 if (gaprow == null || gaprow.annotations == null
429 || gaprow.annotations.length < endCol)
432 * called with a bad alignment annotation row
433 * wait for it to be initialised properly
437 // always set ranges again
438 gaprow.graphMax = nseq;
440 double scale = 0.8 / nseq;
441 for (int i = startCol; i < endCol; i++)
443 ProfileI profile = profiles.get(i);
447 * happens if sequences calculated over were
448 * shorter than alignment width
450 gaprow.annotations[i] = null;
454 final int gapped = profile.getNonGapped();
456 String description = "" + gapped;
458 gaprow.annotations[i] = new Annotation("", description, '\0', gapped,
459 jalview.util.ColorUtils.bleachColour(Color.DARK_GRAY,
460 (float) scale * gapped));
465 * Returns a tooltip showing either
467 * <li>the full profile (percentages of all residues present), if
468 * showSequenceLogo is true, or</li>
469 * <li>just the modal (most common) residue(s), if showSequenceLogo is
472 * Percentages are as a fraction of all sequence, or only ungapped sequences
473 * if ignoreGaps is true.
477 * @param showSequenceLogo
480 * the number of decimal places to format percentages to
483 static String getTooltip(ProfileI profile, float pid,
484 boolean showSequenceLogo, boolean ignoreGaps, int dp)
486 ResidueCount counts = profile.getCounts();
488 String description = null;
489 if (counts != null && showSequenceLogo)
491 int normaliseBy = ignoreGaps ? profile.getNonGapped()
492 : profile.getHeight();
493 description = counts.getTooltip(normaliseBy, dp);
497 StringBuilder sb = new StringBuilder(64);
498 String maxRes = profile.getModalResidue();
499 if (maxRes.length() > 1)
501 sb.append("[").append(maxRes).append("]");
507 if (maxRes.length() > 0)
510 Format.appendPercentage(sb, pid, dp);
513 description = sb.toString();
519 * Returns the sorted profile for the given consensus data. The returned array
523 * [profileType, numberOfValues, nonGapCount, charValue1, percentage1, charValue2, percentage2, ...]
524 * in descending order of percentage value
528 * the data object from which to extract and sort values
530 * if true, only non-gapped values are included in percentage
534 public static int[] extractProfile(ProfileI profile, boolean ignoreGaps)
536 int[] rtnval = new int[64];
537 ResidueCount counts = profile.getCounts();
543 SymbolCounts symbolCounts = counts.getSymbolCounts();
544 char[] symbols = symbolCounts.symbols;
545 int[] values = symbolCounts.values;
546 QuickSort.sort(values, symbols);
547 int nextArrayPos = 2;
548 int totalPercentage = 0;
549 final int divisor = ignoreGaps ? profile.getNonGapped()
550 : profile.getHeight();
553 * traverse the arrays in reverse order (highest counts first)
555 for (int i = symbols.length - 1; i >= 0; i--)
557 int theChar = symbols[i];
558 int charCount = values[i];
560 rtnval[nextArrayPos++] = theChar;
561 final int percentage = (charCount * 100) / divisor;
562 rtnval[nextArrayPos++] = percentage;
563 totalPercentage += percentage;
565 rtnval[0] = symbols.length;
566 rtnval[1] = totalPercentage;
567 int[] result = new int[rtnval.length + 1];
568 result[0] = AlignmentAnnotation.SEQUENCE_PROFILE;
569 System.arraycopy(rtnval, 0, result, 1, rtnval.length);
576 * Extract a sorted extract of cDNA codon profile data. The returned array
580 * [profileType, numberOfValues, totalCount, charValue1, percentage1, charValue2, percentage2, ...]
581 * in descending order of percentage value, where the character values encode codon triplets
587 public static int[] extractCdnaProfile(Hashtable hashtable,
590 // this holds #seqs, #ungapped, and then codon count, indexed by encoded
592 int[] codonCounts = (int[]) hashtable.get(PROFILE);
593 int[] sortedCounts = new int[codonCounts.length - 2];
594 System.arraycopy(codonCounts, 2, sortedCounts, 0,
595 codonCounts.length - 2);
597 int[] result = new int[3 + 2 * sortedCounts.length];
598 // first value is just the type of profile data
599 result[0] = AlignmentAnnotation.CDNA_PROFILE;
601 char[] codons = new char[sortedCounts.length];
602 for (int i = 0; i < codons.length; i++)
604 codons[i] = (char) i;
606 QuickSort.sort(sortedCounts, codons);
607 int totalPercentage = 0;
608 int distinctValuesCount = 0;
610 int divisor = ignoreGaps ? codonCounts[1] : codonCounts[0];
611 for (int i = codons.length - 1; i >= 0; i--)
613 final int codonCount = sortedCounts[i];
616 break; // nothing else of interest here
618 distinctValuesCount++;
619 result[j++] = codons[i];
620 final int percentage = codonCount * 100 / divisor;
621 result[j++] = percentage;
622 totalPercentage += percentage;
624 result[2] = totalPercentage;
627 * Just return the non-zero values
629 // todo next value is redundant if we limit the array to non-zero counts
630 result[1] = distinctValuesCount;
631 return Arrays.copyOfRange(result, 0, j);
635 * Compute a consensus for the cDNA coding for a protein alignment.
638 * the protein alignment (which should hold mappings to cDNA
641 * the consensus data stores to be populated (one per column)
643 public static void calculateCdna(AlignmentI alignment,
644 Hashtable[] hconsensus)
646 final char gapCharacter = alignment.getGapCharacter();
647 List<AlignedCodonFrame> mappings = alignment.getCodonFrames();
648 if (mappings == null || mappings.isEmpty())
653 int cols = alignment.getWidth();
654 for (int col = 0; col < cols; col++)
656 // todo would prefer a Java bean for consensus data
657 Hashtable<String, int[]> columnHash = new Hashtable<>();
658 // #seqs, #ungapped seqs, counts indexed by (codon encoded + 1)
659 int[] codonCounts = new int[66];
660 codonCounts[0] = alignment.getSequences().size();
661 int ungappedCount = 0;
662 for (SequenceI seq : alignment.getSequences())
664 if (seq.getCharAt(col) == gapCharacter)
668 List<char[]> codons = MappingUtils.findCodonsFor(seq, col,
670 for (char[] codon : codons)
672 int codonEncoded = CodingUtils.encodeCodon(codon);
673 if (codonEncoded >= 0)
675 codonCounts[codonEncoded + 2]++;
680 codonCounts[1] = ungappedCount;
681 // todo: sort values here, save counts and codons?
682 columnHash.put(PROFILE, codonCounts);
683 hconsensus[col] = columnHash;
688 * Derive displayable cDNA consensus annotation from computed consensus data.
690 * @param consensusAnnotation
691 * the annotation row to be populated for display
692 * @param consensusData
693 * the computed consensus data
694 * @param showProfileLogo
695 * if true show all symbols present at each position, else only the
698 * the number of sequences in the alignment
700 public static void completeCdnaConsensus(
701 AlignmentAnnotation consensusAnnotation,
702 Hashtable[] consensusData, boolean showProfileLogo, int nseqs)
704 if (consensusAnnotation == null
705 || consensusAnnotation.annotations == null
706 || consensusAnnotation.annotations.length < consensusData.length)
708 // called with a bad alignment annotation row - wait for it to be
709 // initialised properly
713 // ensure codon triplet scales with font size
714 consensusAnnotation.scaleColLabel = true;
715 for (int col = 0; col < consensusData.length; col++)
717 Hashtable hci = consensusData[col];
720 // gapped protein column?
723 // array holds #seqs, #ungapped, then codon counts indexed by codon
724 final int[] codonCounts = (int[]) hci.get(PROFILE);
728 * First pass - get total count and find the highest
730 final char[] codons = new char[codonCounts.length - 2];
731 for (int j = 2; j < codonCounts.length; j++)
733 final int codonCount = codonCounts[j];
734 codons[j - 2] = (char) (j - 2);
735 totalCount += codonCount;
739 * Sort array of encoded codons by count ascending - so the modal value
740 * goes to the end; start by copying the count (dropping the first value)
742 int[] sortedCodonCounts = new int[codonCounts.length - 2];
743 System.arraycopy(codonCounts, 2, sortedCodonCounts, 0,
744 codonCounts.length - 2);
745 QuickSort.sort(sortedCodonCounts, codons);
747 int modalCodonEncoded = codons[codons.length - 1];
748 int modalCodonCount = sortedCodonCounts[codons.length - 1];
749 String modalCodon = String
750 .valueOf(CodingUtils.decodeCodon(modalCodonEncoded));
751 if (sortedCodonCounts.length > 1 && sortedCodonCounts[codons.length
752 - 2] == sortedCodonCounts[codons.length - 1])
755 * two or more codons share the modal count
759 float pid = sortedCodonCounts[sortedCodonCounts.length - 1] * 100
760 / (float) totalCount;
763 * todo ? Replace consensus hashtable with sorted arrays of codons and
764 * counts (non-zero only). Include total count in count array [0].
768 * Scan sorted array backwards for most frequent values first. Show
769 * repeated values compactly.
771 StringBuilder mouseOver = new StringBuilder(32);
772 StringBuilder samePercent = new StringBuilder();
773 String percent = null;
774 String lastPercent = null;
775 int percentDecPl = getPercentageDp(nseqs);
777 for (int j = codons.length - 1; j >= 0; j--)
779 int codonCount = sortedCodonCounts[j];
783 * remaining codons are 0% - ignore, but finish off the last one if
786 if (samePercent.length() > 0)
788 mouseOver.append(samePercent).append(": ").append(percent)
793 int codonEncoded = codons[j];
794 final int pct = codonCount * 100 / totalCount;
795 String codon = String
796 .valueOf(CodingUtils.decodeCodon(codonEncoded));
797 StringBuilder sb = new StringBuilder();
798 Format.appendPercentage(sb, pct, percentDecPl);
799 percent = sb.toString();
800 if (showProfileLogo || codonCount == modalCodonCount)
802 if (percent.equals(lastPercent) && j > 0)
804 samePercent.append(samePercent.length() == 0 ? "" : ", ");
805 samePercent.append(codon);
809 if (samePercent.length() > 0)
811 mouseOver.append(samePercent).append(": ").append(lastPercent)
814 samePercent.setLength(0);
815 samePercent.append(codon);
817 lastPercent = percent;
821 consensusAnnotation.annotations[col] = new Annotation(modalCodon,
822 mouseOver.toString(), ' ', pid);
827 * Returns the number of decimal places to show for profile percentages. For
828 * less than 100 sequences, returns zero (the integer percentage value will be
829 * displayed). For 100-999 sequences, returns 1, for 1000-9999 returns 2, etc.
834 protected static int getPercentageDp(long nseq)
846 * Returns the sorted HMM profile for the given column of the alignment. The
847 * returned array contains
850 * [profileType=0, numberOfValues, 100, charValue1, percentage1, charValue2, percentage2, ...]
851 * in descending order of percentage value
856 * @param removeBelowBackground
857 * if true, ignores residues with probability less than their
858 * background frequency
860 * if true, uses the log ratio 'information' measure to scale the
864 public static int[] extractHMMProfile(HiddenMarkovModel hmm, int column,
865 boolean removeBelowBackground, boolean infoHeight)
871 String alphabet = hmm.getSymbols();
872 int size = alphabet.length();
873 char symbols[] = new char[size];
874 int values[] = new int[size];
877 for (int i = 0; i < size; i++)
879 char symbol = alphabet.charAt(i);
881 int value = getAnalogueCount(hmm, column, symbol,
882 removeBelowBackground, infoHeight);
888 * sort symbols by increasing emission probability
890 QuickSort.sort(values, symbols);
892 int[] profile = new int[3 + size * 2];
894 profile[0] = AlignmentAnnotation.SEQUENCE_PROFILE;
899 * order symbol/count profile by decreasing emission probability
904 for (int k = size - 1; k >= 0; k--)
907 int value = values[k];
908 if (removeBelowBackground)
910 percentage = ((float) value) / totalCount * 100f;
914 percentage = value / 100f;
916 int intPercent = Math.round(percentage);
917 profile[arrayPos] = symbols[k];
918 profile[arrayPos + 1] = intPercent;
926 * Converts the emission probability of a residue at a column in the alignment
927 * to a 'count', suitable for rendering as an annotation value
932 * @param removeBelowBackground
933 * if true, returns 0 for any symbol with a match emission
934 * probability less than the background frequency
935 * @infoHeight if true, uses the log ratio 'information content' to scale the
939 static int getAnalogueCount(HiddenMarkovModel hmm, int column,
940 char symbol, boolean removeBelowBackground, boolean infoHeight)
942 double value = hmm.getMatchEmissionProbability(column, symbol);
943 double freq = ResidueProperties.backgroundFrequencies
944 .get(hmm.getAlphabetType()).get(symbol);
945 if (value < freq && removeBelowBackground)
952 value = value * (Math.log(value / freq) / LOG2);
955 value = value * 10000d;
956 return Math.round((float) value);