public void constructMatrix(ArrayList<MiRBaseEntry> mirb, int subnum, String taxo) {
ArrayList<TrainingEntry> pos = positiveData(mirb, subnum);
trainSet_p = parsePara(pos, taxo);
ArrayList<TrainingEntry> neg = negativeData(pos, level);
trainSet_n = parsePara(neg, taxo);
// report
int total = 0;
for (TrainingEntry t : pos) total += t.NumOfMi();
for (TrainingEntry t : neg) total += t.NumOfMi();
System.out.println("Total Training Entries: " + total);
System.out.println("Level: " + level);
}
/**
* parse parameters of all entries
*
* @param in
*/
private svm_problem parsePara(ArrayList<TrainingEntry> in, String taxo) {
int num = 0;
for (TrainingEntry t : in) num += t.NumOfMi();
svm_node[][] data = new svm_node[num][];
int i = 0;
for (TrainingEntry mi : in) {
pri = mi.getPriRNA();
pri.setPriPara();
for (int j = 0; j < mi.NumOfMi(); j++) {
pri.process(mi.getStart(j), mi.getSize(j), mi.getStrand(j));
ps = new ParaSummary(pri, pri.getPreRNA(), pri.getMiRNA());
data[i++] = svm.predictMatrix(mi.getLabel(j), ps.paraMatrix(taxo));
}
}
return svm.trainMatrix(data);
}
public ArrayList<TrainingEntry> positiveData(ArrayList<MiRBaseEntry> milist, int subnum) {
System.out.print("Reads miRBase data ");
ArrayList<TrainingEntry> pos = new ArrayList<TrainingEntry>();
int num = milist.size();
ArrayList<Integer> rmis = randomization(num, num); // randomizate the miRBase entries
MfeFold doFold;
int evNum = 0; // number of experimental verified miRNAs
int uevNum = 0; // number of unexperimental verified miRNAs
int positive = 0; // number of positive mirna
int n = 0;
for (Integer m : rmis) {
if (subnum > 0 && positive >= subnum) break; // get subnum mirans
MiRBaseEntry mi = milist.get(m);
TrainingEntry te = new TrainingEntry();
pri = new PriMiRNA(mi.getPriAccession(), mi.getPriSequence());
// fold pri-miRNA
doFold = new MfeFold(pri.getSeq());
doFold.cal();
pri.setStr(doFold.getStructure());
pri.setEnergy(doFold.getEnergy());
te.setPriRNA(pri); // store pri
int end5 = pri.getStr().lastIndexOf("(") + 1; // 5' end position of hairpin,count from 1
int start3 = pri.getStr().indexOf(")") + 1; // 3' start position of hairpin,count from 1
for (int i = 0; i < mi.getMatNum(); i++) {
// the miRNA should be experimental verified miRNA
if (mi.getEvidence(i).equals("experimental")) {
evNum += 1;
// the miRNA should not have two or more loops
if (end5 >= start3) continue;
int strand;
if (mi.getMatEnd(i) <= end5) strand = 5; // lie in 5'
else if (mi.getMatStart(i) >= start3) strand = 3; // lie in 3'
else continue; // the miRNA should not lie in loop area
te.addMiEntry(mi.getMatStart(i) - 1, mi.getMatEnd(i) - mi.getMatStart(i) + 1, strand, 1);
// for the fasta format mirna data
// te.setMiId(mi.getMatID(i));
// p_fas.put(te.getMiId(te.NumOfMi()-1), te.getMiSeq(te.NumOfMi()-1));
positive++;
} else uevNum += 1;
}
// add positive data
if (te.NumOfMi() > 0) pos.add(te);
n++;
System.out.print(n + ":" + mi.getPriID() + backspace(n + ":" + mi.getPriID()));
}
System.out.println();
// report
System.out.println("Loads " + (evNum + uevNum) + " Entries");
System.out.println("Experimental verified Entries: " + evNum);
System.out.println("Non-experimental verified Entries: " + uevNum);
System.out.println("Positive Entries: " + positive);
return pos;
}
public ArrayList<TrainingEntry> negativeData(ArrayList<TrainingEntry> pos, int level) {
Random r = new Random(); // random number producer
int rS = 0;
ArrayList<TrainingEntry> nes = new ArrayList<TrainingEntry>();
TrainingEntry ne;
int strand = 0;
int negative = 0;
for (TrainingEntry po : pos) {
ne = new TrainingEntry();
ne.setPriRNA(po.getPriRNA());
int end5 = po.getPriRNA().getStr().lastIndexOf("("); // from 0
int start3 = po.getPriRNA().getStr().indexOf(")"); // from 0
int endP5 = end5 + (start3 - end5 - 1) / 2; // 5' endpoint of hairpin,count from 0
int endP3 = start3 - (start3 - end5 - 1) / 2; // 3' endpoint of hairpin, count from 0
int num = po.NumOfMi(); // the number of postive mi
int size = po.getPriRNA().getLength(); // the size of pri in the entry
for (int n = 0; n < level; n++) { // each level
for (int i = 0; i < num; i++) { // each miRNA in the entry
int miL = po.getSize(i); // the positive mi size
HashMap bases = new HashMap();
int flag = 0;
while (bases.size() < size) { // the negative mi start
rS = r.nextInt(size); // get a random start
if (bases.containsKey(rS)) continue;
bases.put(rS, null); // store the positions which have been used
// the random start is at least 5-bp from the true start positions on the entry
int j;
for (j = 0; j < num; j++) {
if (Math.abs(po.getStart(i) - rS) < 5) break;
}
if (j < num) continue;
// if the random start at 5' strand
if ((rS < endP5 && rS + miL - 1 <= endP5)) {
strand = 5;
flag = 1;
break;
}
// if the random start at 3' strand
else if (rS >= endP3 && rS + miL < size) {
strand = 3;
flag = 1;
break;
}
// else continue;
}
if (flag == 1) {
ne.addMiEntry(rS, miL, strand, 0);
// for the fasta format mirna data
// ne.setMiId(po.getMiId(i)+"n"+n);
// n_fas.put(ne.getMiId(i), ne.getMiSeq(i));
negative++;
} else System.out.println("no negative data found on " + po.getPriRNA().getId());
}
}
nes.add(ne);
}
System.out.println("Negative Entries: " + negative);
return nes;
}
