2 * Jalview - A Sequence Alignment Editor and Viewer ($$Version-Rel$$)
3 * Copyright (C) $$Year-Rel$$ The Jalview Authors
5 * This file is part of Jalview.
7 * Jalview is free software: you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation, either version 3
10 * of the License, or (at your option) any later version.
12 * Jalview is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty
14 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR
15 * PURPOSE. See the GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Jalview. If not, see <http://www.gnu.org/licenses/>.
19 * The Jalview Authors are detailed in the 'AUTHORS' file.
21 package jalview.analysis;
23 import java.util.ArrayList;
24 import java.util.BitSet;
25 import java.util.List;
26 import java.util.Vector;
28 import jalview.datamodel.AlignmentAnnotation;
29 import jalview.datamodel.BinaryNode;
30 import jalview.datamodel.ContactListI;
31 import jalview.datamodel.ContactMatrixI;
32 import jalview.math.Matrix;
33 import jalview.viewmodel.AlignmentViewport;
36 * This class implements distance calculations used in constructing a Average
37 * Distance tree (also known as UPGMA)
39 public class AverageDistanceEngine extends TreeEngine
45 AlignmentAnnotation aa;
48 * compute cosine distance matrix for a given contact matrix and create a
53 public AverageDistanceEngine(AlignmentViewport av, AlignmentAnnotation aa,
63 // 0 - normalised dot product
64 // 1 - L1 - ie (abs(v_1-v_2)/dim(v))
65 // L1 is more rational - since can reason about value of difference,
66 // normalised dot product might give cleaner clusters, but more difficult to
71 public void calculate(ContactMatrixI cm)
74 node = new Vector<BinaryNode>();
75 clusters = new Vector<BitSet>();
76 distances = new Matrix(new double[cm.getWidth()][cm.getWidth()]);
77 noseqs = cm.getWidth();
79 double moduli[] = new double[cm.getWidth()];
87 max = cm.getMax() * cm.getMax();
90 for (int i = 0; i < cm.getWidth(); i++)
92 // init the tree engine node for this column
93 BinaryNode cnode = new BinaryNode();
94 cnode.setElement(Integer.valueOf(i));
95 cnode.setName("c" + i);
96 node.addElement(cnode);
97 BitSet bs = new BitSet();
99 clusters.addElement(bs);
101 // compute distance matrix element
102 ContactListI ith = cm.getContactList(i);
104 for (int j = 0; j < i; j++)
106 distances.setValue(i, i, 0);
107 ContactListI jth = cm.getContactList(j);
109 for (int indx = 0; indx < cm.getHeight(); indx++)
115 moduli[i] += ith.getContactAt(indx) * ith.getContactAt(indx);
117 prd += ith.getContactAt(indx) * jth.getContactAt(indx);
122 .abs(ith.getContactAt(indx) - jth.getContactAt(indx));
129 moduli[i] = Math.sqrt(moduli[i]);
131 prd = (moduli[i] != 0 && moduli[j] != 0)
132 ? prd / (moduli[i] * moduli[j])
138 prd /= cm.getHeight();
140 distances.setValue(i, j, prd);
141 distances.setValue(j, i, prd);
145 noClus = clusters.size();
150 * Calculates and saves the distance between the combination of cluster(i) and
151 * cluster(j) and all other clusters. An average of the distances from
152 * cluster(i) and cluster(j) is calculated, weighted by the sizes of each
159 protected void findClusterDistance(int i, int j)
161 int noi = clusters.elementAt(i).cardinality();
162 int noj = clusters.elementAt(j).cardinality();
164 // New distances from cluster i to others
165 double[] newdist = new double[noseqs];
167 for (int l = 0; l < noseqs; l++)
169 if ((l != i) && (l != j))
171 newdist[l] = ((distances.getValue(i, l) * noi)
172 + (distances.getValue(j, l) * noj)) / (noi + noj);
180 for (int ii = 0; ii < noseqs; ii++)
182 distances.setValue(i, ii, newdist[ii]);
183 distances.setValue(ii, i, newdist[ii]);
191 protected double findMinDistance()
193 double min = Double.MAX_VALUE;
195 for (int i = 0; i < (noseqs - 1); i++)
197 for (int j = i + 1; j < noseqs; j++)
199 if (!done.get(i) && !done.get(j))
201 if (distances.getValue(i, j) < min)
206 min = distances.getValue(i, j);
218 protected void findNewDistances(BinaryNode nodei, BinaryNode nodej,
224 BinaryNode sni = nodei;
225 BinaryNode snj = nodej;
230 sni = (BinaryNode) sni.left();
236 snj = (BinaryNode) snj.left();
239 nodei.dist = ((dist / 2) - ih);
240 nodej.dist = ((dist / 2) - jh);
244 * not the right place - OH WELL!
248 * Makes a list of groups, where each group is represented by a node whose
249 * height (distance from the root node), as a fraction of the height of the
250 * whole tree, is greater than the given threshold. This corresponds to
251 * selecting the nodes immediately to the right of a vertical line
252 * partitioning the tree (if the tree is drawn with root to the left). Each
253 * such node represents a group that contains all of the sequences linked to
254 * the child leaf nodes.
259 public List<BinaryNode> groupNodes(float threshold)
261 List<BinaryNode> groups = new ArrayList<BinaryNode>();
262 _groupNodes(groups, getTopNode(), threshold);
266 protected void _groupNodes(List<BinaryNode> groups, BinaryNode nd,
274 if ((nd.height / maxheight) > threshold)
280 _groupNodes(groups, nd.left(), threshold);
281 _groupNodes(groups, nd.right(), threshold);
291 * @return DOCUMENT ME!
293 public double findHeight(BinaryNode nd)
300 if ((nd.left() == null) && (nd.right() == null))
302 nd.height = ((BinaryNode) nd.parent()).height + nd.dist;
304 if (nd.height > maxheight)
315 if (nd.parent() != null)
317 nd.height = ((BinaryNode) nd.parent()).height + nd.dist;
322 nd.height = (float) 0.0;
325 maxheight = findHeight((BinaryNode) (nd.left()));
326 maxheight = findHeight((BinaryNode) (nd.right()));
333 * Search for leaf nodes below (or at) the given node
336 * root node to search from
340 public Vector<BinaryNode> findLeaves(BinaryNode top2)
342 Vector<BinaryNode> leaves = new Vector<BinaryNode>();
343 findLeaves(top2, leaves);
348 * Search for leaf nodes.
351 * root node to search from
353 * Vector of leaves to add leaf node objects too.
355 * @return Vector of leaf nodes on binary tree
357 Vector<BinaryNode> findLeaves(BinaryNode nd, Vector<BinaryNode> leaves)
364 if ((nd.left() == null) && (nd.right() == null)) // Interior node
367 leaves.addElement(nd);
374 * TODO: Identify internal nodes... if (node.isSequenceLabel()) {
375 * leaves.addElement(node); }
377 findLeaves(nd.left(), leaves);
378 findLeaves(nd.right(), leaves);