/**
* Fills the structure (str) string with the predicted structure according to running the CYK
* algorithm and traceback on the provided sequence string (seq). The parameters should be
* double[3] including the following probabilities:
*
* <p>params[0] = p(S->L) params[1] = p(L->s) params[2] = p(F->LS)
*
* @param seq - The sequence to predict from.
* @param str - The string to fill with the predicted value.
* @param params - The Knudson-Hein Grammar parameters as described above.
* @param verbose - Display output to command line if true.
*/
public static String predictKH(String seq, BigDouble[] params, boolean verbose) {
int size = seq.length();
BigDouble[][][] parr = new BigDouble[size][size][3];
int[][][] tau = new int[size][size][3];
String pred = null;
if (size < 1) {
output.out("Invalid sequence provided: \n\t[" + seq + "]");
return pred;
}
output.out("Predicting secondary structure for \n\t\t[" + seq + "]");
for (int i = 0; i < size; i++)
for (int j = 0; j < size; j++) for (int k = 0; k < 3; k++) tau[i][j][k] = -1;
// Fill Array
kh_CYK(seq, params, parr, tau);
BigDouble prob = parr[0][size - 1][0];
output.out("\t-KH Maximum Probability for sequence is " + prob);
// Trace Back
if (prob.compareTo(0) > 0) {
pred = kh_trace_back(tau);
output.out("\t-KH most likely structure for sequence is\n\t\t[ " + pred + " ]");
}
return pred;
}
public static String predictFromGrammar23S(
String seq, String nat, Grammar grammar, boolean verbose) {
int size = seq.length();
Map<String, BigDouble> parr = new HashMap<String, BigDouble>();
int[][][] tau = new int[size][size][3];
output.out("\t\tTau allocated");
String pred = null;
if (size < 1) {
output.out("\nInvalid sequence provided: \n\t[ " + seq + " ]");
return pred;
}
output.out("\nPredicting secondary structure for \n\t\t[ " + seq + " ]");
long start = System.currentTimeMillis();
for (int i = 0; i < size; i++) {
// Progress Bar
CommandLine.DisplayBar(seq.length(), i, ((long) (System.currentTimeMillis() - start) / 1000));
for (int j = i; j < size; j++) {
String ij = i + ":" + (i + j);
for (int k = 0; k < 3; k++) {
tau[i][j][k] = -1;
parr.put(ij + ":" + k, new BigDouble(0));
}
}
}
CommandLine.DisplayBarFinish();
// Fill Array
if (grammar instanceof PfoldGrammar) {
PfoldGrammar pfold = (PfoldGrammar) grammar;
// _23S_CYK(seq, pfold.getKH_params(), pfold
// .getPfold_paramsUnmatched(), pfold
// .getPfold_paramsBasePairs(), parr, tau);
}
BigDouble prob = parr.get("0:" + (size - 1) + ":" + 0);
output.out("\t-Probability of highest probability parse for sequence is \n\t\t" + prob);
if (prob != null) {
// grammar.recordProbability(prob);
// Trace Back
if (prob.compareTo(0) > 0) {
pred = kh_trace_back(tau);
{
output.out("\t-Highest probability parse generates structure\n\t\t[ " + pred + " ]");
if (nat != null)
output.out(
"\t-FMeasure for predicted structure is\n\t\t" + Compare.getFMeasureBD(nat, pred));
}
}
}
return pred;
}
/**
* Runs the CYK algorithm on the provided sequence with the provided parameters. The int array
* should be array size int[n-1][n-1][3] where "n" is the length of the sequence.
*
* <p>The maximum probability for the provided sequence can be found at index arr[0][n-1][0]
*
* @param seq - The sequence to calculate the maximum probability.
* @param params - The Knudson-Hein grammar probabilities.
* @param arr - The probability array to fill for the algorithm.
* @param arr - The traceback array to fill for the algorithm.
*/
private static void kh_CYK(String seq, BigDouble[] p, BigDouble[][][] arr, int[][][] tau) {
int S = 0, L = 1, F = 2;
BigDouble temp, tempProd;
BigDouble S_LS, L_dFd, F_dFd;
S_LS = new BigDouble(p[S]);
L_dFd = new BigDouble(p[L]);
F_dFd = new BigDouble(p[F]);
for (int i = 0; i < seq.length(); i++) {
if (seq.charAt(i) == 's') {
arr[i][i][L] = p[L];
tau[i][i][S] = 0;
tau[i][i][L] = 0;
tau[i][i][F] = 0;
}
}
for (int j = 0; j < seq.length(); j++) {
for (int i = 0; i + j < seq.length(); i++) {
int ij = i + j;
if (seq.charAt(i) == 'd' && seq.charAt(ij) == 'd' && j > 2) {
// ///////
// L->dFd
temp = L_dFd.mult(arr[i + 1][ij - 1][F]);
if (arr[i][ij][L].compareTo(temp) <= 0) {
arr[i][ij][L] = temp;
tau[i][ij][L] = (F << 8) | 0xff;
}
// ///////
// F->dFd
temp = F_dFd.mult(arr[i + 1][ij - 1][F]);
if (arr[i][ij][F].compareTo(temp) <= 0) {
arr[i][ij][F] = temp;
tau[i][ij][F] = (F << 8) | 0xff;
}
}
// /////
// S->L
temp = p[S].mult(arr[i][ij][L]);
if (arr[i][ij][S].compareTo(temp) <= 0) {
arr[i][ij][S] = temp;
tau[i][ij][S] = (L << 8) | 0xff;
}
for (int k = i; k < ij; k++) {
tempProd = arr[i][k][L].mult(arr[k + 1][ij][S]);
// //////
// S->LS
temp = S_LS.mult(tempProd);
if (arr[i][ij][S].compareTo(temp) <= 0) {
arr[i][ij][S] = temp;
tau[i][ij][S] = (k << 16) | (L << 8) | (S);
}
// //////
// F->LS
temp = p[F].mult(tempProd);
if (arr[i][ij][F].compareTo(temp) <= 0) {
arr[i][ij][F] = temp;
tau[i][ij][F] = (k << 16) | (L << 8) | (S);
}
}
}
}
}
