private void removeOverlap(List<ClusteredResult> clusteredResultList) {
Collections.sort(clusteredResultList);
Collections.reverse(clusteredResultList);
for (int i = clusteredResultList.size() - 1; i >= 1; i--) {
for (int j = i - 1; j >= 0; j--)
for (OptMapResultNode result1 : clusteredResultList.get(i).updatedResult)
for (OptMapResultNode result2 : clusteredResultList.get(j).updatedResult)
if (result1.overlap(result2)) {
clusteredResultList.remove(i);
break;
}
}
}
/**
* Process the confidence, a measure of uniqueness of score
*
* @param crList
*/
private void processConfidence(List<ClusteredResult> crList) {
int itemCount = confItemCount;
if (itemCount == -1 || itemCount > crList.size()) itemCount = crList.size();
double minscore = Double.MAX_VALUE;
double bestscore = Double.MIN_VALUE;
for (int i = 0; i < itemCount; i++) {
ClusteredResult cr = crList.get(i);
if (cr.score < minscore) minscore = cr.score;
if (cr.score > bestscore) bestscore = cr.score;
}
int base = 1;
double total = 0;
if (minscore == bestscore)
// evenly distributed, rare case
for (int i = 0; i < itemCount; i++) {
ClusteredResult cr = crList.get(i);
for (OptMapResultNode result : cr.updatedResult)
result.confidence = (1 / (double) crList.size());
}
else {
for (int i = 0; i < itemCount; i++) {
ClusteredResult cr = crList.get(i);
total += (cr.score - minscore) + base;
}
for (int i = 0; i < itemCount; i++) {
ClusteredResult cr = crList.get(i);
for (OptMapResultNode result : cr.updatedResult)
result.confidence = ((cr.score - minscore + base) / (double) total);
}
}
// Confidence not calculated within itemCount range is set as 0
for (int i = itemCount; i < crList.size(); i++) {
ClusteredResult cr = crList.get(i);
for (OptMapResultNode result : cr.updatedResult) result.confidence = 0;
}
}
/**
* Check if two partial maps can be connected. Trimming is done on overlapping partial maps
*
* <p>It is essential to use trimming. Yet, there are two drawbacks using trimming 1. Trimming is
* expensive. 2. Trimming may lead to inappropriate results (This is done by validation using
* trimear)
*
* @param optrefmap
* @param maxTrim
* @param result1
* @param result2
* @param vmProcessor
* @param ear
* @return
*/
private TrimResult trimOverlap(
LinkedHashMap<String, DataNode> optrefmap,
OptMapResultNode[] trim1Result,
OptMapResultNode[] trim2Result,
PathBuilderFilter pbFilter,
VirtualMapProcessor vmProcessor) {
int penalty = vmProcessor.calcBasicPenalty(trim1Result[0], trim2Result[0]);
double bestscore = Double.NEGATIVE_INFINITY;
int bestTrim1 = -1;
int bestTrim2 = -1;
boolean[][] trimPair = new boolean[maxTrim + 1][maxTrim + 1];
// temp increasing maxTrim greatly:
// 1. if the two score addition is already lower than any one of the score, discard
// 2. maxTrim
for (int trimmed = 0; trimmed <= maxTrim; trimmed++) {
for (int trim1 = 0; trim1 <= trimmed; trim1++) {
int trim2 = trimmed - trim1;
if (trimPair[trim1][trim2]) // this trim pair should no longer be done
continue;
if ((trim1Result[0].cigar.getMatch() - trim1 >= minMatch)
&& (trim2Result[0].cigar.getMatch() - trim2 >= minMatch)) {
OptMapResultNode tresult1 = trim1Result[trim1];
OptMapResultNode tresult2 = trim2Result[trim2];
// trim1Result[trim1 - 1] must exist
if (tresult1 == null) {
tresult1 = new OptMapResultNode(trim1Result[trim1 - 1]);
tresult1.trimResult(-1 * trim1Result[trim1 - 1].mappedstrand, optrefmap);
tresult1.updateScore(optrefmap, match, fpp, fnp);
trim1Result[trim1] = tresult1;
}
if (tresult2 == null) {
tresult2 = new OptMapResultNode(trim2Result[trim2 - 1]);
tresult2.trimResult(1 * trim2Result[trim2 - 1].mappedstrand, optrefmap);
tresult2.updateScore(optrefmap, match, fpp, fnp);
trim2Result[trim2] = tresult2;
}
boolean withinTrimear = true;
withinTrimear =
Math.abs(tresult1.getMapScale() - 1) < trimear
&& Math.abs(tresult2.getMapScale() - 1) < trimear;
boolean removeLaterTrim = false;
if (pbFilter.checkPass(tresult1, tresult2)) {
if (withinTrimear && !tresult1.overlap(tresult2)) {
if (tresult1.mappedscore
+ tresult2.mappedscore
- vmProcessor.calcPenalty(tresult1, tresult2)
> bestscore) {
bestscore =
tresult1.mappedscore
+ tresult2.mappedscore
- vmProcessor.calcPenalty(tresult1, tresult2);
bestTrim1 = trim1;
bestTrim2 = trim2;
}
removeLaterTrim = true;
} else if (withinTrimear
&& tresult1.mappedscore + tresult2.mappedscore - penalty
< bestscore) // We could not calcPenalty directly when
// tresult1.overlap(tresult2)
removeLaterTrim = true;
} else removeLaterTrim = true;
if (removeLaterTrim)
for (int i = trim1; i <= maxTrim; i++)
for (int j = trim2; j <= maxTrim; j++) trimPair[i][j] = true;
}
}
}
if (bestTrim1 == -1) return TrimResult.FailedTrimResult();
else return TrimResult.SuccessfulTrimResult(trim1Result[bestTrim1], trim2Result[bestTrim2]);
}
/**
* Cluster the results by using 1. Path builder filter which limits indel, inversion and
* translocation 2. vmProcessor to calculate the score
*
* <p>The logic includes two steps 1. Grouping 2. Build paths between partial maps (which may
* undergo trimming) 3. Link paths to paths (not partial maps to partial maps) 4. Get the path
* with best score and convert it back to partial maps
*
* @param mapList
* @param pbFilter
* @param vmProcessor
* @return
*/
private List<ClusteredResult> cluster(
List<OptMapResultNode> mapList, PathBuilderFilter pbFilter, VirtualMapProcessor vmProcessor) {
// exit if no result is input
if (mapList == null || mapList.size() == 0) return new ArrayList<ClusteredResult>();
// Pre-update the results
for (OptMapResultNode result : mapList) vmProcessor.calcScore(result);
List<ClusteredResult> clusteredResultList = new ArrayList<ClusteredResult>();
// Grouping close results to be 1. clustered effectively 2. output one clustered result per
// group
List<List<OptMapResultNode>> groupedMapList =
group(mapList, pbFilter.sameStrand, pbFilter.closeReference, pbFilter.closeFragment);
// In each result group
for (List<OptMapResultNode> groupedMap : groupedMapList) {
// Build a path whenever the results can be joined
// Please see ClusterPathNode for details
List<ClusterPathNode> clusterPathList = buildPath(groupedMap, pbFilter, vmProcessor);
// Link the paths
linkPath(clusterPathList, vmProcessor);
// Get the best path and extract the final result maps
List<OptMapResultNode> clusteredGroupMap =
convertPathToMapList(getBestPath(clusterPathList), vmProcessor);
// if (false)
// {
// for (ClusterPathNode cp : clusterPathList) {
// if (cp.currentScore < 0)
// continue;
// if (cp.secondMap != null)
// continue;
// List<ClusterPathNode> bestPath = new ArrayList<ClusterPathNode>();
// ClusterPathNode recentPath = cp;
// while (recentPath != null)
// {
// bestPath.add(recentPath);
// recentPath = recentPath.previousPath;
// }
// Collections.reverse(bestPath);
// if (convertPathToMapList(bestPath).isEmpty())
// continue;
// System.out.println("PATH - " + cp.currentScore);
// for (OptMapResultNode result : convertPathToMapList(bestPath))
// System.out.println(result);
// }
// }
ClusteredResult cr = new ClusteredResult();
// We want to retain the original alignments
List<OptMapResultNode> expandedGroupedMap =
new ArrayList<OptMapResultNode>(); // Saving the original results
for (OptMapResultNode map : groupedMap) expandedGroupedMap.addAll(map.getRealMap());
// We want to import the modified alignments
List<OptMapResultNode> expandedClusterGroupMap =
new ArrayList<OptMapResultNode>(); // Saving the updated results
for (OptMapResultNode map : clusteredGroupMap)
expandedClusterGroupMap.addAll(map.getRealMap());
// checkDirectLink(expandedClusterGroupMap, vmProcessor);
cr.importUpdatedResult(expandedClusterGroupMap);
if (cr.updatedResult.size() > 0) {
cr.process(vmProcessor);
clusteredResultList.add(cr);
}
}
return clusteredResultList;
}
/**
* Group the partial maps
*
* @param mapList
* @param strandSpecific
* @param closeReference
* @param closeFragment
* @return
*/
public List<List<OptMapResultNode>> group(
List<OptMapResultNode> mapList,
boolean strandSpecific,
long closeReference,
long closeFragment) {
if (strandSpecific) {
List<OptMapResultNode> forwardList = new ArrayList<OptMapResultNode>();
List<OptMapResultNode> reverseList = new ArrayList<OptMapResultNode>();
List<List<OptMapResultNode>> unknownListList =
new ArrayList<List<OptMapResultNode>>(); // for virtual fragment
for (OptMapResultNode map : mapList) {
if (map.mappedstrand == 1) forwardList.add(map);
else if (map.mappedstrand == -1) reverseList.add(map);
else {
List<OptMapResultNode> unknownList = new ArrayList<OptMapResultNode>();
unknownList.add(map);
unknownListList.add(unknownList);
}
}
List<List<OptMapResultNode>> clusteredList = new ArrayList<List<OptMapResultNode>>();
clusteredList.addAll(group(forwardList, false, closeReference, closeFragment));
clusteredList.addAll(group(reverseList, false, closeReference, closeFragment));
clusteredList.addAll(unknownListList);
return clusteredList;
} else {
// first group according to reference
List<List<OptMapResultNode>> clusteredList = new ArrayList<List<OptMapResultNode>>();
Collections.sort(mapList, OptMapResultNode.mappedstartcomparator);
List<List<OptMapResultNode>> groupRefList = new ArrayList<List<OptMapResultNode>>();
if (closeReference < 0) clusteredList.add(mapList);
else {
GenomicPosNode region = null;
List<OptMapResultNode> recentList = null;
for (OptMapResultNode map : mapList) {
if (region == null
|| map.mappedRegion.start == -1
|| !map.isClose(region, closeReference)) {
recentList = new ArrayList<OptMapResultNode>();
groupRefList.add(recentList);
if (map.mappedRegion.start == -1) region = null;
else
region =
new GenomicPosNode(
map.mappedRegion.ref, map.mappedRegion.start, map.mappedRegion.stop);
}
recentList.add(map);
if (map.mappedRegion.stop > region.stop)
region = new GenomicPosNode(region.ref, region.start, map.mappedRegion.stop);
}
clusteredList = groupRefList;
}
if (closeFragment < 0) return clusteredList;
else {
List<List<OptMapResultNode>> groupFragList = new ArrayList<List<OptMapResultNode>>();
for (List<OptMapResultNode> tmpMapList : clusteredList) {
Collections.sort(tmpMapList, OptMapResultNode.subfragstartstopcomparator);
SimpleLongLocation loc = null;
List<OptMapResultNode> recentList = null;
for (OptMapResultNode map : tmpMapList) {
SimpleLongLocation maploc =
new SimpleLongLocation(
map.length(0, map.subfragstart - 1), map.length(0, map.subfragstop + 1));
if (loc == null || (!loc.overlap(maploc, closeFragment))) {
recentList = new ArrayList<OptMapResultNode>();
groupFragList.add(recentList);
loc = maploc;
}
recentList.add(map);
if (maploc.max > loc.max) loc.max = maploc.max;
if (maploc.min < loc.min) loc.min = maploc.min;
}
}
clusteredList = groupFragList;
return clusteredList;
}
}
}
