/**
* Print an alignment map in a concise representation. Edges are given as two numbers separated by
* '>'. They are chained together where possible, or separated by spaces where disjoint or
* branched.
*
* <p>Note that more concise representations may be possible. Examples:
* <li>1>2>3>1
* <li>1>2>3>2 4>3
*
* @param alignment The input function, as a map (see {@link
* AlignmentTools#alignmentAsMap(AFPChain)})
* @param identity An identity-like function providing the isomorphism between the codomain of
* alignment (of type <T>) and the domain (type <S>).
* @return
*/
public static <S, T> String toConciseAlignmentString(Map<S, T> alignment, Map<T, S> identity) {
// Clone input to prevent changes
Map<S, T> alig = new HashMap<S, T>(alignment);
// Generate inverse alignment
Map<S, List<S>> inverse = new HashMap<S, List<S>>();
for (Entry<S, T> e : alig.entrySet()) {
S val = identity.get(e.getValue());
if (inverse.containsKey(val)) {
List<S> l = inverse.get(val);
l.add(e.getKey());
} else {
List<S> l = new ArrayList<S>();
l.add(e.getKey());
inverse.put(val, l);
}
}
StringBuilder str = new StringBuilder();
while (!alig.isEmpty()) {
// Pick an edge and work upstream to a root or cycle
S seedNode = alig.keySet().iterator().next();
S node = seedNode;
if (inverse.containsKey(seedNode)) {
node = inverse.get(seedNode).iterator().next();
while (node != seedNode && inverse.containsKey(node)) {
node = inverse.get(node).iterator().next();
}
}
// Now work downstream, deleting edges as we go
seedNode = node;
str.append(node);
while (alig.containsKey(node)) {
S lastNode = node;
node = identity.get(alig.get(lastNode));
// Output
str.append('>');
str.append(node);
// Remove edge
alig.remove(lastNode);
List<S> inv = inverse.get(node);
if (inv.size() > 1) {
inv.remove(node);
} else {
inverse.remove(node);
}
}
if (!alig.isEmpty()) {
str.append(' ');
}
}
return str.toString();
}
/**
* Takes a protein sequence string with capital and lowercase letters and sets its {@link
* ProteinSequence#getUserCollection() user collection} to record which letters are uppercase
* (aligned) and which are lowercase (unaligned).
*
* @param sequence Make sure <em>not</em> to use {@link ProteinSequence#getSequenceAsString()} for
* this, as it won't preserve upper- and lower-case
*/
public static List<Object> getAlignedUserCollection(String sequence) {
List<Object> aligned = new ArrayList<Object>(sequence.length());
for (char c : sequence.toCharArray()) {
aligned.add(Character.isUpperCase(c));
}
return aligned;
}
/**
* identify additional groups that are not directly attached to amino acids.
*
* @param mc {@link ModifiedCompound}.
* @param chain a {@link Chain}.
* @return a list of added groups.
*/
private void identifyAdditionalAttachments(
ModifiedCompound mc, List<Group> ligands, Map<String, Chain> mapChainIdChain) {
if (ligands.isEmpty()) {
return;
}
// TODO: should the additional groups only be allowed to the identified
// ligands or both amino acids and ligands? Currently only on ligands
// ligands to amino acid bonds for same modification of unknown category
// will be combined in mergeModComps()
// TODO: how about chain-chain links?
List<Group> identifiedGroups = new ArrayList<Group>();
for (StructureGroup num : mc.getGroups(false)) {
Group group;
try {
// String numIns = "" + num.getResidueNumber();
// if (num.getInsCode() != null) {
// numIns += num.getInsCode();
// }
ResidueNumber resNum = new ResidueNumber();
resNum.setChainId(num.getChainId());
resNum.setSeqNum(num.getResidueNumber());
resNum.setInsCode(num.getInsCode());
// group = chain.getGroupByPDB(numIns);
group = mapChainIdChain.get(num.getChainId()).getGroupByPDB(resNum);
} catch (StructureException e) {
logger.error("Exception: ", e);
// should not happen
continue;
}
identifiedGroups.add(group);
}
int start = 0;
int n = identifiedGroups.size();
while (n > start) {
for (Group group1 : ligands) {
for (int i = start; i < n; i++) {
Group group2 = identifiedGroups.get(i);
if (!identifiedGroups.contains(group1)) {
List<Atom[]> linkedAtoms =
StructureUtil.findAtomLinkages(group1, group2, false, bondLengthTolerance);
if (!linkedAtoms.isEmpty()) {
for (Atom[] atoms : linkedAtoms) {
mc.addAtomLinkage(
StructureUtil.getStructureAtomLinkage(atoms[0], false, atoms[1], false));
}
identifiedGroups.add(group1);
break;
}
}
}
}
start = n;
n = identifiedGroups.size();
}
}
/**
* Takes an AFPChain and replaces the optimal alignment based on an alignment map
*
* <p>Parameters are filled with defaults (often null) or sometimes calculated.
*
* <p>For a way to create a new AFPChain, see {@link AlignmentTools#createAFPChain(Atom[], Atom[],
* ResidueNumber[], ResidueNumber[])}
*
* @param afpChain The alignment to be modified
* @param alignment The new alignment, as a Map
* @throws StructureException if an error occurred during superposition
* @see AlignmentTools#createAFPChain(Atom[], Atom[], ResidueNumber[], ResidueNumber[])
*/
public static AFPChain replaceOptAln(
AFPChain afpChain, Atom[] ca1, Atom[] ca2, Map<Integer, Integer> alignment)
throws StructureException {
// Determine block lengths
// Sort ca1 indices, then start a new block whenever ca2 indices aren't
// increasing monotonically.
Integer[] res1 = alignment.keySet().toArray(new Integer[0]);
Arrays.sort(res1);
List<Integer> blockLens = new ArrayList<Integer>(2);
int optLength = 0;
Integer lastRes = alignment.get(res1[0]);
int blkLen = lastRes == null ? 0 : 1;
for (int i = 1; i < res1.length; i++) {
Integer currRes = alignment.get(res1[i]); // res2 index
assert (currRes
!= null); // could be converted to if statement if assertion doesn't hold; just modify
// below as well.
if (lastRes < currRes) {
blkLen++;
} else {
// CP!
blockLens.add(blkLen);
optLength += blkLen;
blkLen = 1;
}
lastRes = currRes;
}
blockLens.add(blkLen);
optLength += blkLen;
// Create array structure for alignment
int[][][] optAln = new int[blockLens.size()][][];
int pos1 = 0; // index into res1
for (int blk = 0; blk < blockLens.size(); blk++) {
optAln[blk] = new int[2][];
blkLen = blockLens.get(blk);
optAln[blk][0] = new int[blkLen];
optAln[blk][1] = new int[blkLen];
int pos = 0; // index into optAln
while (pos < blkLen) {
optAln[blk][0][pos] = res1[pos1];
Integer currRes = alignment.get(res1[pos1]);
optAln[blk][1][pos] = currRes;
pos++;
pos1++;
}
}
assert (pos1 == optLength);
// Create length array
int[] optLens = new int[blockLens.size()];
for (int i = 0; i < blockLens.size(); i++) {
optLens[i] = blockLens.get(i);
}
return replaceOptAln(afpChain, ca1, ca2, blockLens.size(), optLens, optAln);
}
/**
* Builds an {@link AFPChain} from already-matched arrays of atoms and residues.
*
* @param ca1 An array of atoms in the first structure
* @param ca2 An array of atoms in the second structure
* @param residues1 An array of {@link ResidueNumber ResidueNumbers} in the first structure that
* are aligned. Only null ResidueNumbers are considered to be unaligned
* @param residues2 An array of {@link ResidueNumber ResidueNumbers} in the second structure that
* are aligned. Only null ResidueNumbers are considered to be unaligned
* @throws StructureException
*/
private static AFPChain buildAlignment(
Atom[] ca1, Atom[] ca2, ResidueNumber[] residues1, ResidueNumber[] residues2)
throws StructureException {
// remove any gap
// this includes the ones introduced by the nullifying above
List<ResidueNumber> alignedResiduesList1 = new ArrayList<ResidueNumber>();
List<ResidueNumber> alignedResiduesList2 = new ArrayList<ResidueNumber>();
for (int i = 0; i < residues1.length; i++) {
if (residues1[i] != null && residues2[i] != null) {
alignedResiduesList1.add(residues1[i]);
alignedResiduesList2.add(residues2[i]);
}
}
ResidueNumber[] alignedResidues1 =
alignedResiduesList1.toArray(new ResidueNumber[alignedResiduesList1.size()]);
ResidueNumber[] alignedResidues2 =
alignedResiduesList2.toArray(new ResidueNumber[alignedResiduesList2.size()]);
AFPChain afpChain = AlignmentTools.createAFPChain(ca1, ca2, alignedResidues1, alignedResidues2);
afpChain.setAlgorithmName("unknown");
AlignmentTools.updateSuperposition(afpChain, ca1, ca2);
afpChain.setBlockSize(new int[] {afpChain.getNrEQR()});
afpChain.setBlockRmsd(new double[] {afpChain.getTotalRmsdOpt()});
afpChain.setBlockGap(new int[] {afpChain.getGapLen()});
return afpChain;
}
private void processCrosslink1(
Map<Component, Set<Group>> mapCompGroups,
List<ModifiedCompound> modComps,
ProteinModification mod,
List<Component> components) {
// modified residue
// TODO: is this the correct logic for CROSS_LINK_1?
Set<Group> modifiedResidues = mapCompGroups.get(components.get(0));
if (modifiedResidues != null) {
for (Group residue : modifiedResidues) {
StructureGroup strucGroup = StructureUtil.getStructureGroup(residue, true);
ModifiedCompound modRes = new ModifiedCompoundImpl(mod, strucGroup);
modComps.add(modRes);
}
}
}
/**
* Assembly the matched linkages.
*
* @param matchedAtomsOfLinkages
* @param mod
* @param condition
* @param ret ModifiedCompound will be stored here.
*/
private void assembleLinkages(
List<List<Atom[]>> matchedAtomsOfLinkages,
ProteinModification mod,
List<ModifiedCompound> ret) {
ModificationCondition condition = mod.getCondition();
List<ModificationLinkage> modLinks = condition.getLinkages();
int nLink = matchedAtomsOfLinkages.size();
int[] indices = new int[nLink];
Set<ModifiedCompound> identifiedCompounds = new HashSet<ModifiedCompound>();
while (indices[0] < matchedAtomsOfLinkages.get(0).size()) {
List<Atom[]> atomLinkages = new ArrayList<Atom[]>(nLink);
for (int iLink = 0; iLink < nLink; iLink++) {
Atom[] atoms = matchedAtomsOfLinkages.get(iLink).get(indices[iLink]);
atomLinkages.add(atoms);
}
if (matchLinkages(modLinks, atomLinkages)) {
// matched
int n = atomLinkages.size();
List<StructureAtomLinkage> linkages = new ArrayList<StructureAtomLinkage>(n);
for (int i = 0; i < n; i++) {
Atom[] linkage = atomLinkages.get(i);
StructureAtomLinkage link =
StructureUtil.getStructureAtomLinkage(
linkage[0], residues.contains(linkage[0].getGroup()),
linkage[1], residues.contains(linkage[1].getGroup()));
linkages.add(link);
}
ModifiedCompound mc = new ModifiedCompoundImpl(mod, linkages);
if (!identifiedCompounds.contains(mc)) {
ret.add(mc);
identifiedCompounds.add(mc);
}
}
// indices++ (e.g. [0,0,1]=>[0,0,2]=>[1,2,0])
int i = nLink - 1;
while (i >= 0) {
if (i == 0 || indices[i] < matchedAtomsOfLinkages.get(i).size() - 1) {
indices[i]++;
break;
} else {
indices[i] = 0;
i--;
}
}
}
}
/**
* Retrieves the optimum alignment from an AFPChain and returns it as a java collection. The
* result is indexed in the same way as {@link AFPChain#getOptAln()}, but has the correct size().
*
* <pre>
* List<List<List<Integer>>> aln = getOptAlnAsList(AFPChain afpChain);
* aln.get(blockNum).get(structureNum={0,1}).get(pos)</pre>
*
* @param afpChain
* @return
*/
public static List<List<List<Integer>>> getOptAlnAsList(AFPChain afpChain) {
int[][][] optAln = afpChain.getOptAln();
int[] optLen = afpChain.getOptLen();
List<List<List<Integer>>> blocks = new ArrayList<List<List<Integer>>>(afpChain.getBlockNum());
for (int blockNum = 0; blockNum < afpChain.getBlockNum(); blockNum++) {
// TODO could improve speed an memory by wrapping the arrays with
// an unmodifiable list, similar to Arrays.asList(...) but with the
// correct size parameter.
List<Integer> align1 = new ArrayList<Integer>(optLen[blockNum]);
List<Integer> align2 = new ArrayList<Integer>(optLen[blockNum]);
for (int pos = 0; pos < optLen[blockNum]; pos++) {
align1.add(optAln[blockNum][0][pos]);
align2.add(optAln[blockNum][1][pos]);
}
List<List<Integer>> block = new ArrayList<List<Integer>>(2);
block.add(align1);
block.add(align2);
blocks.add(block);
}
return blocks;
}
/** Merge identified modified compounds if linked. */
private void mergeModComps(List<ModifiedCompound> modComps) {
TreeSet<Integer> remove = new TreeSet<Integer>();
int n = modComps.size();
for (int icurr = 1; icurr < n; icurr++) {
ModifiedCompound curr = modComps.get(icurr);
String id = curr.getModification().getId();
if (ProteinModificationRegistry.getById(id).getCategory() != ModificationCategory.UNDEFINED)
continue;
// find linked compounds that before curr
// List<Integer> merging = new ArrayList<Integer>();
int ipre = 0;
for (; ipre < icurr; ipre++) {
if (remove.contains(ipre)) continue;
ModifiedCompound pre = modComps.get(ipre);
if (!Collections.disjoint(pre.getGroups(false), curr.getGroups(false))) {
break;
}
}
if (ipre < icurr) {
ModifiedCompound mcKeep = modComps.get(ipre);
// merge modifications of the same type
if (mcKeep.getModification().getId().equals(id)) {
// merging the current one to the previous one
mcKeep.addAtomLinkages(curr.getAtomLinkages());
remove.add(icurr);
}
}
}
Iterator<Integer> it = remove.descendingIterator();
while (it.hasNext()) {
modComps.remove(it.next().intValue());
}
}
/**
* Applies an alignment k times. Eg if alignmentMap defines function f(x), this returns a function
* f^k(x)=f(f(...f(x)...)).
*
* <p>To allow for functions with different domains and codomains, the identity function allows
* converting back in a reasonable way. For instance, if alignmentMap represented an alignment
* between two proteins with different numbering schemes, the identity function could calculate
* the offset between residue numbers, eg I(x) = x-offset.
*
* <p>When an identity function is provided, the returned function calculates f^k(x) = f(I( f(I(
* ... f(x) ... )) )).
*
* @param <S>
* @param <T>
* @param alignmentMap The input function, as a map (see {@link
* AlignmentTools#alignmentAsMap(AFPChain)})
* @param identity An identity-like function providing the isomorphism between the codomain of
* alignmentMap (of type <T>) and the domain (type <S>).
* @param k The number of times to apply the alignment
* @return A new alignment. If the input function is not automorphic (one-to-one), then some
* inputs may map to null, indicating that the function is undefined for that input.
*/
public static <S, T> Map<S, T> applyAlignment(Map<S, T> alignmentMap, Map<T, S> identity, int k) {
// This implementation simply applies the map k times.
// If k were large, it would be more efficient to do this recursively,
// (eg f^4 = (f^2)^2) but k will usually be small.
if (k < 0) throw new IllegalArgumentException("k must be positive");
if (k == 1) {
return new HashMap<S, T>(alignmentMap);
}
// Convert to lists to establish a fixed order
List<S> preimage = new ArrayList<S>(alignmentMap.keySet()); // currently unmodified
List<S> image = new ArrayList<S>(preimage);
for (int n = 1; n < k; n++) {
// apply alignment
for (int i = 0; i < image.size(); i++) {
S pre = image.get(i);
T intermediate = (pre == null ? null : alignmentMap.get(pre));
S post = (intermediate == null ? null : identity.get(intermediate));
image.set(i, post);
}
}
Map<S, T> imageMap = new HashMap<S, T>(alignmentMap.size());
// TODO handle nulls consistently.
// assure that all the residues in the domain are valid keys
/*
for(int i=0;i<preimage.size();i++) {
S pre = preimage.get(i);
T intermediate = (pre==null?null: alignmentMap.get(pre));
S post = (intermediate==null?null: identity.get(intermediate));
imageMap.put(post, null);
}
*/
// now populate with actual values
for (int i = 0; i < preimage.size(); i++) {
S pre = preimage.get(i);
// image is currently f^k-1(x), so take the final step
S preK1 = image.get(i);
T postK = (preK1 == null ? null : alignmentMap.get(preK1));
imageMap.put(pre, postK);
}
return imageMap;
}
/**
* Record unidentifiable atom linkages in a chain. Only linkages between two residues or one
* residue and one ligand will be recorded.
*/
private void recordUnidentifiableAtomLinkages(
List<ModifiedCompound> modComps, List<Group> ligands) {
// first put identified linkages in a map for fast query
Set<StructureAtomLinkage> identifiedLinkages = new HashSet<StructureAtomLinkage>();
for (ModifiedCompound mc : modComps) {
identifiedLinkages.addAll(mc.getAtomLinkages());
}
// record
// cross link
int nRes = residues.size();
for (int i = 0; i < nRes - 1; i++) {
Group group1 = residues.get(i);
for (int j = i + 1; j < nRes; j++) {
Group group2 = residues.get(j);
List<Atom[]> linkages =
StructureUtil.findAtomLinkages(group1, group2, true, bondLengthTolerance);
for (Atom[] atoms : linkages) {
StructureAtomLinkage link =
StructureUtil.getStructureAtomLinkage(atoms[0], true, atoms[1], true);
unidentifiableAtomLinkages.add(link);
}
}
}
// attachment
int nLig = ligands.size();
for (int i = 0; i < nRes; i++) {
Group group1 = residues.get(i);
for (int j = 0; j < nLig; j++) {
Group group2 = ligands.get(j);
if (group1.equals(group2)) { // overlap between residues and ligands
continue;
}
List<Atom[]> linkages =
StructureUtil.findAtomLinkages(group1, group2, false, bondLengthTolerance);
for (Atom[] atoms : linkages) {
StructureAtomLinkage link =
StructureUtil.getStructureAtomLinkage(atoms[0], true, atoms[1], false);
unidentifiableAtomLinkages.add(link);
}
}
}
}
/**
* @param linkages
* @param atomLinkages
* @return true if atomLinkages satisfy the condition; false, otherwise.
*/
private boolean matchLinkages(List<ModificationLinkage> linkages, List<Atom[]> atomLinkages) {
int nLink = linkages.size();
if (nLink != atomLinkages.size()) {
return false;
}
for (int i = 0; i < nLink - 1; i++) {
ModificationLinkage link1 = linkages.get(i);
Atom[] atoms1 = atomLinkages.get(i);
for (int j = i + 1; j < nLink; j++) {
ModificationLinkage link2 = linkages.get(j);
Atom[] atoms2 = atomLinkages.get(j);
// check components
if (((link1.getIndexOfComponent1() == link2.getIndexOfComponent1())
!= (atoms1[0].getGroup().equals(atoms2[0].getGroup())))
|| ((link1.getIndexOfComponent1() == link2.getIndexOfComponent2())
!= (atoms1[0].getGroup().equals(atoms2[1].getGroup())))
|| ((link1.getIndexOfComponent2() == link2.getIndexOfComponent1())
!= (atoms1[1].getGroup().equals(atoms2[0].getGroup())))
|| ((link1.getIndexOfComponent2() == link2.getIndexOfComponent2())
!= (atoms1[1].getGroup().equals(atoms2[1].getGroup())))) {
return false;
}
// check atoms
String label11 = link1.getLabelOfAtomOnComponent1();
String label12 = link1.getLabelOfAtomOnComponent2();
String label21 = link2.getLabelOfAtomOnComponent1();
String label22 = link2.getLabelOfAtomOnComponent2();
if ((label11 != null && label21 != null && label11.equals(label21))
!= (atoms1[0].equals(atoms2[0]))
|| (label11 != null && label22 != null && label11.equals(label22))
!= (atoms1[0].equals(atoms2[1]))
|| (label12 != null && label21 != null && label12.equals(label21))
!= (atoms1[1].equals(atoms2[0]))
|| (label12 != null && label22 != null && label12.equals(label22))
!= (atoms1[1].equals(atoms2[1]))) {
return false;
}
}
}
return true;
}
/**
* Uses two sequences each with a corresponding structure to create an AFPChain corresponding to
* the alignment. Provided only for convenience since FastaReaders return such maps.
*
* @param sequences A Map containing exactly two entries from sequence names as Strings to gapped
* ProteinSequences; the name is ignored
* @see #fastaToAfpChain(ProteinSequence, ProteinSequence, Structure, Structure)
* @throws StructureException
*/
public static AFPChain fastaToAfpChain(
Map<String, ProteinSequence> sequences, Structure structure1, Structure structure2)
throws StructureException {
if (sequences.size() != 2) {
throw new IllegalArgumentException(
"There must be exactly 2 sequences, but there were " + sequences.size());
}
if (structure1 == null || structure2 == null) {
throw new IllegalArgumentException("A structure is null");
}
List<ProteinSequence> seqs = new ArrayList<ProteinSequence>();
List<String> names = new ArrayList<String>(2);
for (Map.Entry<String, ProteinSequence> entry : sequences.entrySet()) {
seqs.add(entry.getValue());
names.add(entry.getKey());
}
return fastaToAfpChain(seqs.get(0), seqs.get(1), structure1, structure2);
}
/**
* Identify a set of modifications in a a list of chains.
*
* @param chains query {@link Chain}s.
* @param potentialModifications query {@link ProteinModification}s.
*/
public void identify(
final List<Chain> chains, final Set<ProteinModification> potentialModifications) {
if (chains == null) {
throw new IllegalArgumentException("Null structure.");
}
if (potentialModifications == null) {
throw new IllegalArgumentException("Null potentialModifications.");
}
reset();
if (potentialModifications.isEmpty()) {
return;
}
Map<String, Chain> mapChainIdChain = new HashMap<String, Chain>(chains.size());
residues = new ArrayList<Group>();
List<Group> ligands = new ArrayList<Group>();
Map<Component, Set<Group>> mapCompGroups = new HashMap<Component, Set<Group>>();
for (Chain chain : chains) {
mapChainIdChain.put(chain.getChainID(), chain);
List<Group> ress = StructureUtil.getAminoAcids(chain);
// List<Group> ligs = chain.getAtomLigands();
List<Group> ligs = StructureTools.filterLigands(chain.getAtomGroups());
residues.addAll(ress);
residues.removeAll(ligs);
ligands.addAll(ligs);
addModificationGroups(potentialModifications, ress, ligs, mapCompGroups);
}
if (residues.isEmpty()) {
String pdbId = "?";
if (chains.size() > 0) {
Structure struc = chains.get(0).getParent();
if (struc != null) pdbId = struc.getPDBCode();
}
logger.warn(
"No amino acids found for {}. Either you did not parse the PDB file with alignSEQRES records, or this record does not contain any amino acids.",
pdbId);
}
List<ModifiedCompound> modComps = new ArrayList<ModifiedCompound>();
for (ProteinModification mod : potentialModifications) {
ModificationCondition condition = mod.getCondition();
List<Component> components = condition.getComponents();
if (!mapCompGroups.keySet().containsAll(components)) {
// not all components exist for this mod.
continue;
}
int sizeComps = components.size();
if (sizeComps == 1) {
processCrosslink1(mapCompGroups, modComps, mod, components);
} else {
processMultiCrosslink(mapCompGroups, modComps, mod, condition);
}
}
if (recordAdditionalAttachments) {
// identify additional groups that are not directly attached to amino acids.
for (ModifiedCompound mc : modComps) {
identifyAdditionalAttachments(mc, ligands, mapChainIdChain);
}
}
mergeModComps(modComps);
identifiedModifiedCompounds.addAll(modComps);
// record unidentifiable linkage
if (recordUnidentifiableModifiedCompounds) {
recordUnidentifiableAtomLinkages(modComps, ligands);
recordUnidentifiableModifiedResidues(modComps);
}
}
/**
* @param modifications a set of {@link ProteinModification}s.
* @param residues
* @param ligands
* @param saveTo save result to
* @return map from component to list of corresponding residues in the chain.
*/
private void addModificationGroups(
final Set<ProteinModification> modifications,
final List<Group> residues,
final List<Group> ligands,
final Map<Component, Set<Group>> saveTo) {
if (residues == null || ligands == null || modifications == null) {
throw new IllegalArgumentException("Null argument(s).");
}
Map<Component, Set<Component>> mapSingleMultiComps = new HashMap<Component, Set<Component>>();
for (ProteinModification mod : modifications) {
ModificationCondition condition = mod.getCondition();
for (Component comp : condition.getComponents()) {
for (String pdbccId : comp.getPdbccIds()) {
Component single =
Component.of(Collections.singleton(pdbccId), comp.isNTerminal(), comp.isCTerminal());
Set<Component> mult = mapSingleMultiComps.get(single);
if (mult == null) {
mult = new HashSet<Component>();
mapSingleMultiComps.put(single, mult);
}
mult.add(comp);
}
}
}
{
// ligands
Set<Component> ligandsWildCard = mapSingleMultiComps.get(Component.of("*"));
for (Group group : ligands) {
String pdbccId = group.getPDBName().trim();
Set<Component> comps = mapSingleMultiComps.get(Component.of(pdbccId));
for (Component comp : unionComponentSet(ligandsWildCard, comps)) {
Set<Group> gs = saveTo.get(comp);
if (gs == null) {
gs = new LinkedHashSet<Group>();
saveTo.put(comp, gs);
}
gs.add(group);
}
}
}
{
// residues
if (residues.isEmpty()) {
return;
}
Set<Component> residuesWildCard = mapSingleMultiComps.get(Component.of("*"));
// for all residues
for (Group group : residues) {
String pdbccId = group.getPDBName().trim();
Set<Component> comps = mapSingleMultiComps.get(Component.of(pdbccId));
for (Component comp : unionComponentSet(residuesWildCard, comps)) {
Set<Group> gs = saveTo.get(comp);
if (gs == null) {
gs = new LinkedHashSet<Group>();
saveTo.put(comp, gs);
}
gs.add(group);
}
}
// for N-terminal
int nRes = residues.size();
int iRes = 0;
Group res;
do {
// for all ligands on N terminal and the first residue
res = residues.get(iRes++);
Set<Component> nTermWildCard = mapSingleMultiComps.get(Component.of("*", true, false));
Set<Component> comps = mapSingleMultiComps.get(Component.of(res.getPDBName(), true, false));
for (Component comp : unionComponentSet(nTermWildCard, comps)) {
Set<Group> gs = saveTo.get(comp);
if (gs == null) {
gs = new LinkedHashSet<Group>();
saveTo.put(comp, gs);
}
gs.add(res);
}
} while (iRes < nRes && ligands.contains(res));
// for C-terminal
iRes = residues.size() - 1;
do {
// for all ligands on C terminal and the last residue
res = residues.get(iRes--);
Set<Component> cTermWildCard = mapSingleMultiComps.get(Component.of("*", false, true));
Set<Component> comps = mapSingleMultiComps.get(Component.of(res.getPDBName(), false, true));
for (Component comp : unionComponentSet(cTermWildCard, comps)) {
Set<Group> gs = saveTo.get(comp);
if (gs == null) {
gs = new LinkedHashSet<Group>();
saveTo.put(comp, gs);
}
gs.add(res);
}
} while (iRes >= 0 && ligands.contains(res));
}
}
/** Get matched atoms for all linkages. */
private List<List<Atom[]>> getMatchedAtomsOfLinkages(
ModificationCondition condition, Map<Component, Set<Group>> mapCompGroups) {
List<ModificationLinkage> linkages = condition.getLinkages();
int nLink = linkages.size();
List<List<Atom[]>> matchedAtomsOfLinkages = new ArrayList<List<Atom[]>>(nLink);
for (int iLink = 0; iLink < nLink; iLink++) {
ModificationLinkage linkage = linkages.get(iLink);
Component comp1 = linkage.getComponent1();
Component comp2 = linkage.getComponent2();
// boolean isAA1 = comp1.;
// boolean isAA2 = comp2.getType()==true;
Set<Group> groups1 = mapCompGroups.get(comp1);
Set<Group> groups2 = mapCompGroups.get(comp2);
List<Atom[]> list = new ArrayList<Atom[]>();
List<String> potentialNamesOfAtomOnGroup1 = linkage.getPDBNameOfPotentialAtomsOnComponent1();
for (String name : potentialNamesOfAtomOnGroup1) {
if (name.equals("*")) {
// wildcard
potentialNamesOfAtomOnGroup1 = null; // search all atoms
break;
}
}
List<String> potentialNamesOfAtomOnGroup2 = linkage.getPDBNameOfPotentialAtomsOnComponent2();
for (String name : potentialNamesOfAtomOnGroup2) {
if (name.equals("*")) {
// wildcard
potentialNamesOfAtomOnGroup2 = null; // search all atoms
break;
}
}
for (Group g1 : groups1) {
for (Group g2 : groups2) {
if (g1.equals(g2)) {
continue;
}
// only for wildcard match of two residues
boolean ignoreNCLinkage =
potentialNamesOfAtomOnGroup1 == null
&& potentialNamesOfAtomOnGroup2 == null
&& residues.contains(g1)
&& residues.contains(g2);
Atom[] atoms =
StructureUtil.findNearestAtomLinkage(
g1,
g2,
potentialNamesOfAtomOnGroup1,
potentialNamesOfAtomOnGroup2,
ignoreNCLinkage,
bondLengthTolerance);
if (atoms != null) {
list.add(atoms);
}
}
}
if (list.isEmpty()) {
// broken linkage
break;
}
matchedAtomsOfLinkages.add(list);
}
return matchedAtomsOfLinkages;
}
/**
* @param a
* @param ca1
* @param ca2
* @return
* @throws StructureException if an error occurred during superposition
*/
public static AFPChain splitBlocksByTopology(AFPChain a, Atom[] ca1, Atom[] ca2)
throws StructureException {
int[][][] optAln = a.getOptAln();
int blockNum = a.getBlockNum();
int[] optLen = a.getOptLen();
// Determine block lengths
// Split blocks if residue indices don't increase monotonically
List<Integer> newBlkLen = new ArrayList<Integer>();
boolean blockChanged = false;
for (int blk = 0; blk < blockNum; blk++) {
int currLen = 1;
for (int pos = 1; pos < optLen[blk]; pos++) {
if (optAln[blk][0][pos] <= optAln[blk][0][pos - 1]
|| optAln[blk][1][pos] <= optAln[blk][1][pos - 1]) {
// start a new block
newBlkLen.add(currLen);
currLen = 0;
blockChanged = true;
}
currLen++;
}
if (optLen[blk] < 2) {
newBlkLen.add(optLen[blk]);
} else {
newBlkLen.add(currLen);
}
}
// Check if anything needs to be split
if (!blockChanged) {
return a;
}
// Split blocks
List<int[][]> blocks = new ArrayList<int[][]>(newBlkLen.size());
int oldBlk = 0;
int pos = 0;
for (int blkLen : newBlkLen) {
if (blkLen == optLen[oldBlk]) {
assert (pos == 0); // should be the whole block
// Use the old block
blocks.add(optAln[oldBlk]);
} else {
int[][] newBlock = new int[2][blkLen];
assert (pos + blkLen <= optLen[oldBlk]); // don't overrun block
for (int i = 0; i < blkLen; i++) {
newBlock[0][i] = optAln[oldBlk][0][pos + i];
newBlock[1][i] = optAln[oldBlk][1][pos + i];
}
pos += blkLen;
blocks.add(newBlock);
if (pos == optLen[oldBlk]) {
// Finished this oldBlk, start the next
oldBlk++;
pos = 0;
}
}
}
// Store new blocks
int[][][] newOptAln = blocks.toArray(new int[0][][]);
int[] newBlockLens = new int[newBlkLen.size()];
for (int i = 0; i < newBlkLen.size(); i++) {
newBlockLens[i] = newBlkLen.get(i);
}
return replaceOptAln(a, ca1, ca2, blocks.size(), newBlockLens, newOptAln);
}
/**
* Tries to detect symmetry in an alignment.
*
* <p>Conceptually, an alignment is a function f:A->B between two sets of integers. The function
* may have simple topology (meaning that if two elements of A are close, then their images in B
* will also be close), or may have more complex topology (such as a circular permutation). This
* function checks <i>alignment</i> against a reference function <i>identity</i>, which should
* have simple topology. It then tries to determine the symmetry order of <i>alignment</i>
* relative to <i>identity</i>, up to a maximum order of <i>maxSymmetry</i>.
*
* <p><strong>Details</strong><br>
* Considers the offset (in number of residues) which a residue moves after undergoing <i>n</i>
* alternating transforms by alignment and identity. If <i>n</i> corresponds to the intrinsic
* order of the alignment, this will be small. This algorithm tries increasing values of <i>n</i>
* and looks for abrupt decreases in the root mean squared offset. If none are found at
* <i>n</i><=maxSymmetry, the alignment is reported as non-symmetric.
*
* @param alignment The alignment to test for symmetry
* @param identity An alignment with simple topology which approximates the sequential
* relationship between the two proteins. Should map in the reverse direction from alignment.
* @param maxSymmetry Maximum symmetry to consider. High values increase the calculation time and
* can lead to overfitting.
* @param minimumMetricChange Percent decrease in root mean squared offsets in order to declare
* symmetry. 0.4f seems to work well for CeSymm.
* @return The order of symmetry of alignment, or 1 if no order <= maxSymmetry is found.
* @see IdentityMap For a simple identity function
*/
public static int getSymmetryOrder(
Map<Integer, Integer> alignment,
Map<Integer, Integer> identity,
final int maxSymmetry,
final float minimumMetricChange) {
List<Integer> preimage = new ArrayList<Integer>(alignment.keySet()); // currently unmodified
List<Integer> image = new ArrayList<Integer>(preimage);
int bestSymmetry = 1;
double bestMetric = Double.POSITIVE_INFINITY; // lower is better
boolean foundSymmetry = false;
if (debug) {
logger.trace("Symm\tPos\tDelta");
}
for (int n = 1; n <= maxSymmetry; n++) {
int deltasSq = 0;
int numDeltas = 0;
// apply alignment
for (int i = 0; i < image.size(); i++) {
Integer pre = image.get(i);
Integer intermediate = (pre == null ? null : alignment.get(pre));
Integer post = (intermediate == null ? null : identity.get(intermediate));
image.set(i, post);
if (post != null) {
int delta = post - preimage.get(i);
deltasSq += delta * delta;
numDeltas++;
if (debug) {
logger.debug("%d\t%d\t%d\n", n, preimage.get(i), delta);
}
}
}
// Metrics: RMS compensates for the trend of smaller numDeltas with higher order
// Not normalizing by numDeltas favors smaller orders
double metric = Math.sqrt((double) deltasSq / numDeltas); // root mean squared distance
if (!foundSymmetry && metric < bestMetric * minimumMetricChange) {
// n = 1 is never the best symmetry
if (bestMetric < Double.POSITIVE_INFINITY) {
foundSymmetry = true;
}
bestSymmetry = n;
bestMetric = metric;
}
// When debugging need to loop over everything. Unneeded in production
if (!debug && foundSymmetry) {
break;
}
}
if (foundSymmetry) {
return bestSymmetry;
} else {
return 1;
}
}
