package jalview.datamodel;
+import jalview.io.HMMFile;
+import jalview.schemes.ResidueProperties;
+import jalview.util.Comparison;
+import jalview.util.MapList;
+
+import java.util.ArrayList;
+import java.util.Arrays;
import java.util.HashMap;
+import java.util.List;
import java.util.Map;
/**
- * Data structure to hold a HMM file
- */
-/**
+ * Data structure which stores a hidden Markov model
+ *
* @author TZVanaalten
*
*/
public class HiddenMarkovModel
{
+ private static final char GAP_DASH = '-';
+
+ public final static String YES = "yes";
+
+ public final static String NO = "no";
+
+ public static final int MATCHTOMATCH = 0;
+
+ public static final int MATCHTOINSERT = 1;
+
+ public static final int MATCHTODELETE = 2;
+
+ public static final int INSERTTOMATCH = 3;
+
+ public static final int INSERTTOINSERT = 4;
+
+ public static final int DELETETOMATCH = 5;
+
+ public static final int DELETETODELETE = 6;
- // Stores file properties
+ private static final double LOG2 = Math.log(2);
+
+ /*
+ * properties read from HMM file header lines
+ */
private Map<String, String> fileProperties = new HashMap<>();
- public String getAccessionNumber()
+ private String fileHeader;
+
+ /*
+ * the symbols used in this model e.g. "ACGT"
+ */
+ private String alphabet;
+
+ /*
+ * symbol lookup index into the alphabet for 'A' to 'Z'
+ */
+ private int[] symbolIndexLookup = new int['Z' - 'A' + 1];
+
+ /*
+ * Nodes in the model. The begin node is at index 0, and contains
+ * average emission probabilities for each symbol.
+ */
+ private List<HMMNode> nodes = new ArrayList<>();
+
+ /*
+ * the aligned HMM consensus sequence extracted from the HMM profile
+ */
+ private SequenceI hmmSeq;
+
+ /*
+ * mapping from HMM nodes to residues of the hmm consensus sequence
+ */
+ private Mapping mapToHmmConsensus;
+
+ // stores background frequencies of alignment from which this model came
+ private Map<Character, Float> backgroundFrequencies;
+
+ /**
+ * Constructor
+ */
+ public HiddenMarkovModel()
+ {
+ }
+
+ /**
+ * Copy constructor given a new aligned sequence with which to associate the
+ * HMM profile
+ *
+ * @param hmm
+ * @param sq
+ */
+ public HiddenMarkovModel(HiddenMarkovModel hmm, SequenceI sq)
+ {
+ super();
+ this.fileProperties = new HashMap<>(hmm.fileProperties);
+ this.alphabet = hmm.alphabet;
+ this.nodes = new ArrayList<>(hmm.nodes);
+ this.symbolIndexLookup = hmm.symbolIndexLookup;
+ this.fileHeader = new String(hmm.fileHeader);
+ this.hmmSeq = sq;
+ this.backgroundFrequencies = hmm.getBackgroundFrequencies();
+ if (sq.getDatasetSequence() == hmm.mapToHmmConsensus.getTo())
+ {
+ // same dataset sequence e.g. after realigning search results
+ this.mapToHmmConsensus = hmm.mapToHmmConsensus;
+ }
+ else
+ {
+ // different dataset sequence e.g. after loading HMM from project
+ this.mapToHmmConsensus = new Mapping(sq.getDatasetSequence(),
+ hmm.mapToHmmConsensus.getMap());
+ }
+ }
+
+ /**
+ * Returns the information content at a specified column, calculated as the
+ * sum (over possible symbols) of the log ratio
+ *
+ * <pre>
+ * log(emission probability / background probability) / log(2)
+ * </pre>
+ *
+ * @param column
+ * column position (base 0)
+ * @return
+ */
+ public float getInformationContent(int column)
{
- return fileProperties.get("ACC");
+ float informationContent = 0f;
+
+ for (char symbol : getSymbols().toCharArray())
+ {
+ float freq = ResidueProperties.backgroundFrequencies
+ .get(getAlphabetType()).get(symbol);
+ float prob = (float) getMatchEmissionProbability(column, symbol);
+ informationContent += prob * Math.log(prob / freq);
+ }
+
+ informationContent = informationContent / (float) LOG2;
+
+ return informationContent;
}
- public String getDescription()
+ /**
+ * Gets the file header of the .hmm file this model came from
+ *
+ * @return
+ */
+ public String getFileHeader()
{
- return fileProperties.get("DESC");
+ return fileHeader;
}
- public int getModelLength()
+ /**
+ * Sets the file header of this model.
+ *
+ * @param header
+ */
+ public void setFileHeader(String header)
{
- return Integer.parseInt(fileProperties.get("LENG"));
+ fileHeader = header;
}
- public int getMaxInstanceLength()
+ /**
+ * Returns the symbols used in this hidden Markov model
+ *
+ * @return
+ */
+ public String getSymbols()
+ {
+ return alphabet;
+ }
+
+ /**
+ * Gets the node in the hidden Markov model at the specified position.
+ *
+ * @param nodeIndex
+ * The index of the node requested. Node 0 optionally contains the
+ * average match emission probabilities across the entire model, and
+ * always contains the insert emission probabilities and state
+ * transition probabilities for the begin node. Node 1 contains the
+ * first node in the HMM that can correspond to a column in the
+ * alignment.
+ * @return
+ */
+ public HMMNode getNode(int nodeIndex)
{
- return Integer.parseInt(fileProperties.get("MAXL"));
+ return nodes.get(nodeIndex);
}
- // gets type of symbol alphabet - "amino", "DNA", "RNA"
- public String getAlphabetType()
+ /**
+ * Returns the name of the sequence alignment on which the HMM is based.
+ *
+ * @return
+ */
+ public String getName()
+ {
+ return fileProperties.get(HMMFile.NAME);
+ }
+
+ /**
+ * Answers the string value of the property (parsed from an HMM file) for the
+ * given key, or null if the property is not present
+ *
+ * @param key
+ * @return
+ */
+ public String getProperty(String key)
+ {
+ return fileProperties.get(key);
+ }
+
+ /**
+ * Answers true if the property with the given key is present with a value of
+ * "yes" (not case-sensitive), else false
+ *
+ * @param key
+ * @return
+ */
+ public boolean getBooleanProperty(String key)
{
- return fileProperties.get("ALPH");
+ return YES.equalsIgnoreCase(fileProperties.get(key));
}
- // returns boolean indicating whether the reference annotation character field
- // for each match state is valid or ignored
- public boolean getReferenceAnnotationFlag()
+ /**
+ * Returns the length of the hidden Markov model. The value returned is the
+ * LENG property if specified, else the number of nodes, excluding the begin
+ * node (which should be the same thing).
+ *
+ * @return
+ */
+ public int getLength()
{
- if (fileProperties.get("RF") == "yes")
+ if (fileProperties.get(HMMFile.LENGTH) == null)
{
- return true;
+ return nodes.size() - 1; // not counting BEGIN node
}
- return false;
+ return Integer.parseInt(fileProperties.get(HMMFile.LENGTH));
}
- // returns boolean indicating whether the model mask annotation character
- // field
- // for each match state is valid or ignored
- public boolean getModelMaskedFlag()
+ /**
+ * Returns the value of mandatory property "ALPH" - "amino", "DNA", "RNA" are
+ * the options. Other alphabets may be added.
+ *
+ * @return
+ */
+ public String getAlphabetType()
{
- if (fileProperties.get("MM") == "yes")
+ return fileProperties.get(HMMFile.ALPHABET);
+ }
+
+ /**
+ * Sets the model alphabet to the symbols in the given string (ignoring any
+ * whitespace), and returns the number of symbols
+ *
+ * @param symbols
+ */
+ public int setAlphabet(String symbols)
+ {
+ String trimmed = symbols.toUpperCase().replaceAll("\\s", "");
+ int count = trimmed.length();
+ alphabet = trimmed;
+ symbolIndexLookup = new int['Z' - 'A' + 1];
+ Arrays.fill(symbolIndexLookup, -1);
+ int ignored = 0;
+
+ /*
+ * save the symbols in order, and a quick lookup of symbol position
+ */
+ for (short i = 0; i < count; i++)
{
- return true;
+ char symbol = trimmed.charAt(i);
+ if (symbol >= 'A' && symbol <= 'Z'
+ && symbolIndexLookup[symbol - 'A'] == -1)
+ {
+ symbolIndexLookup[symbol - 'A'] = i;
+ }
+ else
+ {
+ System.err
+ .println(
+ "Unexpected or duplicated character in HMM ALPHabet: "
+ + symbol);
+ ignored++;
+ }
}
- return false;
+ return count - ignored;
}
- // returns boolean indicating whether the consensus residue field
- // for each match state is valid or ignored
- public boolean getConsensusResidueAnnotationFlag()
+ /**
+ * Answers the node of the model corresponding to an aligned column position
+ * (0...), or null if there is no such node
+ *
+ * @param column
+ * @return
+ */
+ HMMNode getNodeForColumn(int column)
{
- if (fileProperties.get("CONS") == "yes")
+ /*
+ * if the hmm consensus is gapped at the column,
+ * there is no corresponding node
+ */
+ if (Comparison.isGap(hmmSeq.getCharAt(column)))
+ {
+ return null;
+ }
+
+ /*
+ * find the node (if any) that is mapped to the
+ * consensus sequence residue position at the column
+ */
+ int seqPos = hmmSeq.findPosition(column);
+ int[] nodeNo = mapToHmmConsensus.getMap().locateInFrom(seqPos, seqPos);
+ if (nodeNo != null)
{
- return true;
+ return getNode(nodeNo[0]);
}
- return false;
+ return null;
}
- // returns boolean indicating whether the consensus structure character field
- // for each match state is valid or ignored
- public boolean getConsensusStructureAnnotationFlag()
+ /**
+ * Gets the match emission probability for a given symbol at a column in the
+ * alignment.
+ *
+ * @param alignColumn
+ * The index of the alignment column, starting at index 0. Index 0
+ * usually corresponds to index 1 in the HMM.
+ * @param symbol
+ * The symbol for which the desired probability is being requested.
+ * @return
+ *
+ */
+ public double getMatchEmissionProbability(int alignColumn, char symbol)
{
- if (fileProperties.get("CS") == "yes")
+ HMMNode node = getNodeForColumn(alignColumn);
+ int symbolIndex = getSymbolIndex(symbol);
+ if (node != null && symbolIndex != -1)
{
- return true;
+ return node.getMatchEmission(symbolIndex);
}
- return false;
+ return 0D;
}
- // returns boolean indicating whether the model mask annotation character
- // field
- // for each match state is valid or ignored
- public boolean getMapAnnotationFlag()
+ /**
+ * Gets the insert emission probability for a given symbol at a column in the
+ * alignment.
+ *
+ * @param alignColumn
+ * The index of the alignment column, starting at index 0. Index 0
+ * usually corresponds to index 1 in the HMM.
+ * @param symbol
+ * The symbol for which the desired probability is being requested.
+ * @return
+ *
+ */
+ public double getInsertEmissionProbability(int alignColumn, char symbol)
+ {
+ HMMNode node = getNodeForColumn(alignColumn);
+ int symbolIndex = getSymbolIndex(symbol);
+ if (node != null && symbolIndex != -1)
+ {
+ return node.getInsertEmission(symbolIndex);
+ }
+ return 0D;
+ }
+
+ /**
+ * Gets the state transition probability for a given symbol at a column in the
+ * alignment.
+ *
+ * @param alignColumn
+ * The index of the alignment column, starting at index 0. Index 0
+ * usually corresponds to index 1 in the HMM.
+ * @param symbol
+ * The symbol for which the desired probability is being requested.
+ * @return
+ *
+ */
+ public double getStateTransitionProbability(int alignColumn,
+ int transition)
{
- if (fileProperties.get("MAP") == "yes")
+ HMMNode node = getNodeForColumn(alignColumn);
+ if (node != null)
{
- return true;
+ return node.getStateTransition(transition);
}
- return false;
+ return 0D;
+ }
+
+ /**
+ * Returns the sequence position linked to the node at the given index. This
+ * corresponds to an aligned column position (counting from 1).
+ *
+ * @param nodeIndex
+ * The index of the node, starting from index 1. Index 0 is the begin
+ * node, which does not correspond to a column in the alignment.
+ * @return
+ */
+ public int getNodeMapPosition(int nodeIndex)
+ {
+ return nodes.get(nodeIndex).getResidueNumber();
+ }
+
+ /**
+ * Returns the consensus residue at the specified node.
+ *
+ * @param nodeIndex
+ * The index of the specified node.
+ * @return
+ */
+ public char getConsensusResidue(int nodeIndex)
+ {
+ char value = nodes.get(nodeIndex).getConsensusResidue();
+ return value;
+ }
+
+ /**
+ * Returns the reference annotation at the specified node.
+ *
+ * @param nodeIndex
+ * The index of the specified node.
+ * @return
+ */
+ public char getReferenceAnnotation(int nodeIndex)
+ {
+ char value = nodes.get(nodeIndex).getReferenceAnnotation();
+ return value;
+ }
+
+ /**
+ * Returns the mask value at the specified node.
+ *
+ * @param nodeIndex
+ * The index of the specified node.
+ * @return
+ */
+ public char getMaskedValue(int nodeIndex)
+ {
+ char value = nodes.get(nodeIndex).getMaskValue();
+ return value;
+ }
+
+ /**
+ * Returns the consensus structure at the specified node.
+ *
+ * @param nodeIndex
+ * The index of the specified node.
+ * @return
+ */
+ public char getConsensusStructure(int nodeIndex)
+ {
+ char value = nodes.get(nodeIndex).getConsensusStructure();
+ return value;
}
+
+ /**
+ * Sets a property read from an HMM file
+ *
+ * @param key
+ * @param value
+ */
+ public void setProperty(String key, String value)
+ {
+ fileProperties.put(key, value);
+ }
+
+ /**
+ * Temporary implementation, should not be used.
+ *
+ * @return
+ */
+ public String getViterbi()
+ {
+ String value;
+ value = fileProperties.get(HMMFile.VITERBI);
+ return value;
+ }
+
+ /**
+ * Temporary implementation, should not be used.
+ *
+ * @return
+ */
+ public String getMSV()
+ {
+ String value;
+ value = fileProperties.get(HMMFile.MSV);
+ return value;
+ }
+
+ /**
+ * Temporary implementation, should not be used.
+ *
+ * @return
+ */
+ public String getForward()
+ {
+ String value;
+ value = fileProperties.get(HMMFile.FORWARD);
+ return value;
+ }
+
+ /**
+ * Constructs the consensus sequence based on the most probable symbol at each
+ * position. Gap characters are inserted for discontinuities in the node map
+ * numbering (if provided), else an ungapped sequence is generated.
+ * <p>
+ * A mapping between the HMM nodes and residue positions of the sequence is
+ * also built and saved.
+ *
+ * @return
+ */
+ void buildConsensusSequence()
+ {
+ List<int[]> toResidues = new ArrayList<>();
+
+ /*
+ * if the HMM provided a map to sequence, use those start/end values,
+ * else just treat it as for a contiguous sequence numbered from 1
+ */
+ boolean hasMap = getBooleanProperty(HMMFile.MAP);
+ int start = hasMap ? getNode(1).getResidueNumber() : 1;
+ int endResNo = hasMap ? getNode(nodes.size() - 1).getResidueNumber()
+ : (start + getLength() - 1);
+ char[] sequence = new char[endResNo];
- // not sure whether to implement this
- // public Date getDate()
- // {
+ int lastResNo = start - 1;
+ int seqOffset = -1;
+ int gapCount = 0;
- // }
- // not sure whether to implement this
- // public String getCommandLineLog()
- // {
+ for (int seqN = 0; seqN < start; seqN++)
+ {
+ sequence[seqN] = GAP_DASH;
+ seqOffset++;
+ }
+
+ for (int nodeNo = 1; nodeNo < nodes.size(); nodeNo++)
+ {
+ HMMNode node = nodes.get(nodeNo);
+ final int resNo = hasMap ? node.getResidueNumber() : lastResNo + 1;
+
+ /*
+ * insert gaps if map numbering is not continuous
+ */
+ while (resNo > lastResNo + 1)
+ {
+ sequence[seqOffset++] = GAP_DASH;
+ lastResNo++;
+ gapCount++;
+ }
+ char consensusResidue = node.getConsensusResidue();
+ if (GAP_DASH == consensusResidue)
+ {
+ /*
+ * no residue annotation in HMM - scan for the symbol
+ * with the highest match emission probability
+ */
+ int symbolIndex = node.getMaxMatchEmissionIndex();
+ consensusResidue = alphabet.charAt(symbolIndex);
+ if (node.getMatchEmission(symbolIndex) < 0.5D)
+ {
+ // follow convention of lower case if match emission prob < 0.5
+ consensusResidue = Character.toLowerCase(consensusResidue);
+ }
+ }
+ sequence[seqOffset++] = consensusResidue;
+ lastResNo = resNo;
+ }
- // }
+ Sequence seq = new Sequence(getName(), sequence, start,
+ lastResNo - gapCount);
+ seq.createDatasetSequence();
+ seq.setHMM(this);
+ this.hmmSeq = seq;
+
+ /*
+ * construct and store Mapping of nodes to residues
+ * note as constructed this is just an identity mapping,
+ * but it allows for greater flexibility in future
+ */
+ List<int[]> fromNodes = new ArrayList<>();
+ fromNodes.add(new int[] { 1, getLength() });
+ toResidues.add(new int[] { seq.getStart(), seq.getEnd() });
+ MapList mapList = new MapList(fromNodes, toResidues, 1, 1);
+ mapToHmmConsensus = new Mapping(seq.getDatasetSequence(), mapList);
+ }
+
+
+ /**
+ * Answers the aligned consensus sequence for the profile. Note this will
+ * return null if called before <code>setNodes</code> has been called.
+ *
+ * @return
+ */
+ public SequenceI getConsensusSequence()
+ {
+ return hmmSeq;
+ }
- // gets the number of sequences that the HMM was trained on
- public int getSequenceNumber()
+ /**
+ * Answers the index position (0...) of the given symbol, or -1 if not a valid
+ * symbol for this HMM
+ *
+ * @param symbol
+ * @return
+ */
+ private int getSymbolIndex(char symbol)
{
- return Integer.parseInt(fileProperties.get("NSEQ"));
+ /*
+ * symbolIndexLookup holds the index for 'A' to 'Z'
+ */
+ char c = Character.toUpperCase(symbol);
+ if ('A' <= c && c <= 'Z')
+ {
+ return symbolIndexLookup[c - 'A'];
+ }
+ return -1;
}
- // gets the effective number determined during sequence weighting
- public int getEffectiveSequenceNumber()
+ /**
+ * Sets the nodes of this HMM, and also extracts the HMM consensus sequence
+ * and a mapping between node numbers and sequence positions
+ *
+ * @param nodeList
+ */
+ public void setNodes(List<HMMNode> nodeList)
{
- return Integer.parseInt(fileProperties.get("EFFN"));
+ nodes = nodeList;
+ if (nodes.size() > 1)
+ {
+ buildConsensusSequence();
+ }
}
- public int getCheckSum()
+ /**
+ * Sets the aligned consensus sequence this HMM is the model for
+ *
+ * @param hmmSeq
+ */
+ public void setHmmSeq(SequenceI hmmSeq)
{
- return Integer.parseInt(fileProperties.get("CKSUM"));
+ this.hmmSeq = hmmSeq;
}
- // need to ask if BigDecimal is best decimal type for this purpose
- // and how to limit number of decimals
- public double getGatheringThresholdGA1()
+ public void setBackgroundFrequencies(Map<Character, Float> bkgdFreqs)
{
- return Double.parseDouble((fileProperties.get("GA1")));
+ backgroundFrequencies = bkgdFreqs;
}
- public void put(String key, String value)
+ public void setBackgroundFrequencies(ResidueCount bkgdFreqs)
{
- fileProperties.put(key, value);
+ backgroundFrequencies = new HashMap<>();
+
+ int total = bkgdFreqs.getTotalResidueCount();
+
+ for (char c : bkgdFreqs.getSymbolCounts().symbols)
+ {
+ backgroundFrequencies.put(c, bkgdFreqs.getCount(c) * 1f / total);
+ }
+
}
+ public Map<Character, Float> getBackgroundFrequencies()
+ {
+ return backgroundFrequencies;
+ }
}
+