001/* 002 * BioJava development code 003 * 004 * This code may be freely distributed and modified under the 005 * terms of the GNU Lesser General Public Licence. This should 006 * be distributed with the code. If you do not have a copy, 007 * see: 008 * 009 * http://www.gnu.org/copyleft/lesser.html 010 * 011 * Copyright for this code is held jointly by the individual 012 * authors. These should be listed in @author doc comments. 013 * 014 * For more information on the BioJava project and its aims, 015 * or to join the biojava-l mailing list, visit the home page 016 * at: 017 * 018 * http://www.biojava.org/ 019 * 020 */ 021package org.biojava.nbio.structure.align.util; 022 023import org.biojava.nbio.structure.*; 024import org.biojava.nbio.structure.align.AFPTwister; 025import org.biojava.nbio.structure.align.ce.CECalculator; 026import org.biojava.nbio.structure.align.fatcat.FatCatFlexible; 027import org.biojava.nbio.structure.align.fatcat.FatCatRigid; 028import org.biojava.nbio.structure.align.model.AFPChain; 029import org.biojava.nbio.structure.align.xml.AFPChainXMLParser; 030import org.biojava.nbio.structure.geometry.Matrices; 031import org.biojava.nbio.structure.geometry.SuperPositions; 032import org.biojava.nbio.structure.jama.Matrix; 033import org.slf4j.Logger; 034import org.slf4j.LoggerFactory; 035 036import java.io.IOException; 037import java.io.Writer; 038import java.util.*; 039import java.util.Map.Entry; 040import java.util.regex.Matcher; 041import java.util.regex.Pattern; 042 043import javax.vecmath.Matrix4d; 044 045/** 046 * Methods for analyzing and manipulating AFPChains and for 047 * other pairwise alignment utilities. <p> 048 * Current methods: replace optimal alignment, create new AFPChain, 049 * format conversion, update superposition, etc. 050 * 051 * @author Spencer Bliven 052 * @author Aleix Lafita 053 * 054 */ 055public class AlignmentTools { 056 057 private static final Logger logger = LoggerFactory.getLogger(AlignmentTools.class); 058 059 060 public static boolean debug = false; 061 062 /** 063 * Checks that the alignment given by afpChain is sequential. This means 064 * that the residue indices of both proteins increase monotonically as 065 * a function of the alignment position (ie both proteins are sorted). 066 * 067 * This will return false for circularly permuted alignments or other 068 * non-topological alignments. It will also return false for cases where 069 * the alignment itself is sequential but it is not stored in the afpChain 070 * in a sorted manner. 071 * 072 * Since algorithms which create non-sequential alignments split the 073 * alignment into multiple blocks, some computational time can be saved 074 * by only checking block boundaries for sequentiality. Setting 075 * <code>checkWithinBlocks</code> to <code>true</code> makes this function slower, 076 * but detects AFPChains with non-sequential blocks. 077 * 078 * Note that this method should give the same results as 079 * {@link AFPChain#isSequentialAlignment()}. However, the AFPChain version 080 * relies on the StructureAlignment algorithm correctly setting this 081 * parameter, which is sadly not always the case. 082 * 083 * @param afpChain An alignment 084 * @param checkWithinBlocks Indicates whether individual blocks should be 085 * checked for sequentiality 086 * @return True if the alignment is sequential. 087 */ 088 public static boolean isSequentialAlignment(AFPChain afpChain, boolean checkWithinBlocks) { 089 int[][][] optAln = afpChain.getOptAln(); 090 int[] alnLen = afpChain.getOptLen(); 091 int blocks = afpChain.getBlockNum(); 092 093 if(blocks < 1) return true; //trivial case 094 if ( alnLen[0] < 1) return true; 095 096 // Check that blocks are sequential 097 if(checkWithinBlocks) { 098 for(int block = 0; block<blocks; block++) { 099 if(alnLen[block] < 1 ) continue; //skip empty blocks 100 101 int prevRes1 = optAln[block][0][0]; 102 int prevRes2 = optAln[block][1][0]; 103 104 for(int pos = 1; pos<alnLen[block]; pos++) { 105 int currRes1 = optAln[block][0][pos]; 106 int currRes2 = optAln[block][1][pos]; 107 108 if(currRes1 < prevRes1) { 109 return false; 110 } 111 if(currRes2 < prevRes2) { 112 return false; 113 } 114 115 prevRes1 = currRes1; 116 prevRes2 = currRes2; 117 } 118 } 119 } 120 121 // Check that blocks are sequential 122 int prevRes1 = optAln[0][0][alnLen[0]-1]; 123 int prevRes2 = optAln[0][1][alnLen[0]-1]; 124 125 for(int block = 1; block<blocks;block++) { 126 if(alnLen[block] < 1 ) continue; //skip empty blocks 127 128 if(optAln[block][0][0]<prevRes1) { 129 return false; 130 } 131 if(optAln[block][1][0]<prevRes2) { 132 return false; 133 } 134 135 prevRes1 = optAln[block][0][alnLen[block]-1]; 136 prevRes2 = optAln[block][1][alnLen[block]-1]; 137 } 138 139 return true; 140 } 141 142 /** 143 * Creates a Map specifying the alignment as a mapping between residue indices 144 * of protein 1 and residue indices of protein 2. 145 * 146 * <p>For example,<pre> 147 * 1234 148 * 5678</pre> 149 * becomes<pre> 150 * 1->5 151 * 2->6 152 * 3->7 153 * 4->8</pre> 154 * 155 * @param afpChain An alignment 156 * @return A mapping from aligned residues of protein 1 to their partners in protein 2. 157 * @throws StructureException If afpChain is not one-to-one 158 */ 159 public static Map<Integer, Integer> alignmentAsMap(AFPChain afpChain) throws StructureException { 160 Map<Integer,Integer> map = new HashMap<>(); 161 162 if( afpChain.getAlnLength() < 1 ) { 163 return map; 164 } 165 int[][][] optAln = afpChain.getOptAln(); 166 int[] optLen = afpChain.getOptLen(); 167 for(int block = 0; block < afpChain.getBlockNum(); block++) { 168 for(int pos = 0; pos < optLen[block]; pos++) { 169 int res1 = optAln[block][0][pos]; 170 int res2 = optAln[block][1][pos]; 171 if(map.containsKey(res1)) { 172 throw new StructureException(String.format("Residue %d aligned to both %d and %d.", res1,map.get(res1),res2)); 173 } 174 map.put(res1,res2); 175 } 176 } 177 return map; 178 } 179 180 /** 181 * Applies an alignment k times. Eg if alignmentMap defines function f(x), 182 * this returns a function f^k(x)=f(f(...f(x)...)). 183 * 184 * @param <T> 185 * @param alignmentMap The input function, as a map (see {@link AlignmentTools#alignmentAsMap(AFPChain)}) 186 * @param k The number of times to apply the alignment 187 * @return A new alignment. If the input function is not automorphic 188 * (one-to-one), then some inputs may map to null, indicating that the 189 * function is undefined for that input. 190 */ 191 public static <T> Map<T,T> applyAlignment(Map<T, T> alignmentMap, int k) { 192 return applyAlignment(alignmentMap, new IdentityMap<T>(), k); 193 } 194 195 /** 196 * Applies an alignment k times. Eg if alignmentMap defines function f(x), 197 * this returns a function f^k(x)=f(f(...f(x)...)). 198 * 199 * To allow for functions with different domains and codomains, the identity 200 * function allows converting back in a reasonable way. For instance, if 201 * alignmentMap represented an alignment between two proteins with different 202 * numbering schemes, the identity function could calculate the offset 203 * between residue numbers, eg I(x) = x-offset. 204 * 205 * When an identity function is provided, the returned function calculates 206 * f^k(x) = f(I( f(I( ... f(x) ... )) )). 207 * 208 * @param <S> 209 * @param <T> 210 * @param alignmentMap The input function, as a map (see {@link AlignmentTools#alignmentAsMap(AFPChain)}) 211 * @param identity An identity-like function providing the isomorphism between 212 * the codomain of alignmentMap (of type T) and the domain (type S). 213 * @param k The number of times to apply the alignment 214 * @return A new alignment. If the input function is not automorphic 215 * (one-to-one), then some inputs may map to null, indicating that the 216 * function is undefined for that input. 217 */ 218 public static <S,T> Map<S,T> applyAlignment(Map<S, T> alignmentMap, Map<T,S> identity, int k) { 219 220 // This implementation simply applies the map k times. 221 // If k were large, it would be more efficient to do this recursively, 222 // (eg f^4 = (f^2)^2) but k will usually be small. 223 224 if(k<0) throw new IllegalArgumentException("k must be positive"); 225 if(k==1) { 226 return new HashMap<>(alignmentMap); 227 } 228 // Convert to lists to establish a fixed order 229 List<S> preimage = new ArrayList<>(alignmentMap.keySet()); // currently unmodified 230 List<S> image = new ArrayList<>(preimage); 231 232 for(int n=1;n<k;n++) { 233 // apply alignment 234 for(int i=0;i<image.size();i++) { 235 S pre = image.get(i); 236 T intermediate = (pre==null?null: alignmentMap.get(pre)); 237 S post = (intermediate==null?null: identity.get(intermediate)); 238 image.set(i, post); 239 } 240 } 241 242 Map<S, T> imageMap = new HashMap<>(alignmentMap.size()); 243 244 //TODO handle nulls consistently. 245 // assure that all the residues in the domain are valid keys 246 /* 247 for(int i=0;i<preimage.size();i++) { 248 S pre = preimage.get(i); 249 T intermediate = (pre==null?null: alignmentMap.get(pre)); 250 S post = (intermediate==null?null: identity.get(intermediate)); 251 imageMap.put(post, null); 252 } 253 */ 254 // now populate with actual values 255 for(int i=0;i<preimage.size();i++) { 256 S pre = preimage.get(i); 257 258 // image is currently f^k-1(x), so take the final step 259 S preK1 = image.get(i); 260 T postK = (preK1==null?null: alignmentMap.get(preK1)); 261 imageMap.put(pre,postK); 262 263 } 264 return imageMap; 265 } 266 267 /** 268 * Helper for {@link #getSymmetryOrder(Map, Map, int, float)} with a true 269 * identity function (X->X). 270 * 271 * <p>This method should only be used in cases where the two proteins 272 * aligned have identical numbering, as for self-alignments. See 273 * {@link #getSymmetryOrder(AFPChain, int, float)} for a way to guess 274 * the sequential correspondence between two proteins. 275 * 276 * @param alignment 277 * @param maxSymmetry 278 * @param minimumMetricChange 279 * @return 280 */ 281 public static int getSymmetryOrder(Map<Integer, Integer> alignment, 282 final int maxSymmetry, final float minimumMetricChange) { 283 return getSymmetryOrder(alignment, new IdentityMap<>(), maxSymmetry, minimumMetricChange); 284 } 285 /** 286 * Tries to detect symmetry in an alignment. 287 * 288 * <p>Conceptually, an alignment is a function f:A->B between two sets of 289 * integers. The function may have simple topology (meaning that if two 290 * elements of A are close, then their images in B will also be close), or 291 * may have more complex topology (such as a circular permutation). This 292 * function checks <i>alignment</i> against a reference function 293 * <i>identity</i>, which should have simple topology. It then tries to 294 * determine the symmetry order of <i>alignment</i> relative to 295 * <i>identity</i>, up to a maximum order of <i>maxSymmetry</i>. 296 * 297 * 298 * <p><strong>Details</strong><br/> 299 * Considers the offset (in number of residues) which a residue moves 300 * after undergoing <i>n</i> alternating transforms by alignment and 301 * identity. If <i>n</i> corresponds to the intrinsic order of the alignment, 302 * this will be small. This algorithm tries increasing values of <i>n</i> 303 * and looks for abrupt decreases in the root mean squared offset. 304 * If none are found at <i>n</i><=maxSymmetry, the alignment is reported as 305 * non-symmetric. 306 * 307 * @param alignment The alignment to test for symmetry 308 * @param identity An alignment with simple topology which approximates 309 * the sequential relationship between the two proteins. Should map in the 310 * reverse direction from alignment. 311 * @param maxSymmetry Maximum symmetry to consider. High values increase 312 * the calculation time and can lead to overfitting. 313 * @param minimumMetricChange Percent decrease in root mean squared offsets 314 * in order to declare symmetry. 0.4f seems to work well for CeSymm. 315 * @return The order of symmetry of alignment, or 1 if no order <= 316 * maxSymmetry is found. 317 * 318 * @see IdentityMap For a simple identity function 319 */ 320 public static int getSymmetryOrder(Map<Integer, Integer> alignment, Map<Integer,Integer> identity, 321 final int maxSymmetry, final float minimumMetricChange) { 322 List<Integer> preimage = new ArrayList<>(alignment.keySet()); // currently unmodified 323 List<Integer> image = new ArrayList<>(preimage); 324 325 int bestSymmetry = 1; 326 double bestMetric = Double.POSITIVE_INFINITY; //lower is better 327 boolean foundSymmetry = false; 328 329 if(debug) { 330 logger.trace("Symm\tPos\tDelta"); 331 } 332 333 for(int n=1;n<=maxSymmetry;n++) { 334 int deltasSq = 0; 335 int numDeltas = 0; 336 // apply alignment 337 for(int i=0;i<image.size();i++) { 338 Integer pre = image.get(i); 339 Integer intermediate = (pre==null?null: alignment.get(pre)); 340 Integer post = (intermediate==null?null: identity.get(intermediate)); 341 image.set(i, post); 342 343 if(post != null) { 344 int delta = post-preimage.get(i); 345 346 deltasSq += delta*delta; 347 numDeltas++; 348 349 if(debug) { 350 logger.debug("%d\t%d\t%d\n",n,preimage.get(i),delta); 351 } 352 } 353 354 } 355 356 // Metrics: RMS compensates for the trend of smaller numDeltas with higher order 357 // Not normalizing by numDeltas favors smaller orders 358 359 double metric = Math.sqrt((double)deltasSq/numDeltas); // root mean squared distance 360 361 if(!foundSymmetry && metric < bestMetric * minimumMetricChange) { 362 // n = 1 is never the best symmetry 363 if(bestMetric < Double.POSITIVE_INFINITY) { 364 foundSymmetry = true; 365 } 366 bestSymmetry = n; 367 bestMetric = metric; 368 } 369 370 // When debugging need to loop over everything. Unneeded in production 371 if(!debug && foundSymmetry) { 372 break; 373 } 374 375 } 376 if(foundSymmetry) { 377 return bestSymmetry; 378 } else { 379 return 1; 380 } 381 } 382 383 384 /** 385 * Guesses the order of symmetry in an alignment 386 * 387 * <p>Uses {@link #getSymmetryOrder(Map alignment, Map identity, int, float)} 388 * to determine the the symmetry order. For the identity alignment, sorts 389 * the aligned residues of each protein sequentially, then defines the ith 390 * residues of each protein to be equivalent. 391 * 392 * <p>Note that the selection of the identity alignment here is <i>very</i> 393 * naive, and only works for proteins with very good coverage. Wherever 394 * possible, it is better to construct an identity function explicitly 395 * from a sequence alignment (or use an {@link IdentityMap} for internally 396 * symmetric proteins) and use {@link #getSymmetryOrder(Map, Map, int, float)}. 397 */ 398 public static int getSymmetryOrder(AFPChain afpChain, int maxSymmetry, float minimumMetricChange) throws StructureException { 399 // alignment comes from the afpChain alignment 400 Map<Integer,Integer> alignment = AlignmentTools.alignmentAsMap(afpChain); 401 402 // Now construct identity to map aligned residues in sequential order 403 Map<Integer, Integer> identity = guessSequentialAlignment(alignment, true); 404 405 406 return AlignmentTools.getSymmetryOrder(alignment, 407 identity, 408 maxSymmetry, minimumMetricChange); 409 } 410 411 /** 412 * Takes a potentially non-sequential alignment and guesses a sequential 413 * version of it. Residues from each structure are sorted sequentially and 414 * then compared directly. 415 * 416 * <p>The results of this method are consistent with what one might expect 417 * from an identity function, and are therefore useful with 418 * {@link #getSymmetryOrder(Map, Map identity, int, float)}. 419 * <ul> 420 * <li>Perfect self-alignments will have the same pre-image and image, 421 * so will map X->X</li> 422 * <li>Gaps and alignment errors will cause errors in the resulting map, 423 * but only locally. Errors do not propagate through the whole 424 * alignment.</li> 425 * </ul> 426 * 427 * <h4>Example:</h4> 428 * A non sequential alignment, represented schematically as 429 * <pre> 430 * 12456789 431 * 78912345</pre> 432 * would result in a map 433 * <pre> 434 * 12456789 435 * 12345789</pre> 436 * @param alignment The non-sequential input alignment 437 * @param inverseAlignment If false, map from structure1 to structure2. If 438 * true, generate the inverse of that map. 439 * @return A mapping from sequential residues of one protein to those of the other 440 * @throws IllegalArgumentException if the input alignment is not one-to-one. 441 */ 442 public static Map<Integer, Integer> guessSequentialAlignment( 443 Map<Integer,Integer> alignment, boolean inverseAlignment) { 444 Map<Integer,Integer> identity = new HashMap<>(); 445 446 SortedSet<Integer> aligned1 = new TreeSet<>(); 447 SortedSet<Integer> aligned2 = new TreeSet<>(); 448 449 for(Entry<Integer,Integer> pair : alignment.entrySet()) { 450 aligned1.add(pair.getKey()); 451 if( !aligned2.add(pair.getValue()) ) 452 throw new IllegalArgumentException("Alignment is not one-to-one for residue "+pair.getValue()+" of the second structure."); 453 } 454 455 Iterator<Integer> it1 = aligned1.iterator(); 456 Iterator<Integer> it2 = aligned2.iterator(); 457 while(it1.hasNext()) { 458 if(inverseAlignment) { // 2->1 459 identity.put(it2.next(),it1.next()); 460 } else { // 1->2 461 identity.put(it1.next(),it2.next()); 462 } 463 } 464 return identity; 465 } 466 467 /** 468 * Retrieves the optimum alignment from an AFPChain and returns it as a 469 * java collection. The result is indexed in the same way as 470 * {@link AFPChain#getOptAln()}, but has the correct size(). 471 * <pre>{@code 472 * List<List<List<Integer>>> aln = getOptAlnAsList(AFPChain afpChain); 473 * aln.get(blockNum).get(structureNum={0,1}).get(pos) 474 * }</pre> 475 * 476 * @param afpChain 477 * @return 478 */ 479 public static List<List<List<Integer>>> getOptAlnAsList(AFPChain afpChain) { 480 int[][][] optAln = afpChain.getOptAln(); 481 int[] optLen = afpChain.getOptLen(); 482 List<List<List<Integer>>> blocks = new ArrayList<>(afpChain.getBlockNum()); 483 for(int blockNum=0;blockNum<afpChain.getBlockNum();blockNum++) { 484 //TODO could improve speed an memory by wrapping the arrays with 485 // an unmodifiable list, similar to Arrays.asList(...) but with the 486 // correct size parameter. 487 List<Integer> align1 = new ArrayList<>(optLen[blockNum]); 488 List<Integer> align2 = new ArrayList<>(optLen[blockNum]); 489 for(int pos=0;pos<optLen[blockNum];pos++) { 490 align1.add(optAln[blockNum][0][pos]); 491 align2.add(optAln[blockNum][1][pos]); 492 } 493 List<List<Integer>> block = new ArrayList<>(2); 494 block.add(align1); 495 block.add(align2); 496 blocks.add(block); 497 } 498 499 return blocks; 500 } 501 502 503 504 /** 505 * A {@code Map<K,V>} can be viewed as a function from K to V. This class represents 506 * the identity function. Getting a value results in the value itself. 507 * 508 * <p>The class is a bit inconsistent when representing its contents. On 509 * the one hand, containsKey(key) is true for all objects. However, 510 * attempting to iterate through the values returns an empty set. 511 * 512 * @author Spencer Bliven 513 * 514 * @param <K> 515 */ 516 public static class IdentityMap<K> extends AbstractMap<K,K> { 517 public IdentityMap() {} 518 519 /** 520 * @param key 521 * @return the key 522 * @throws ClassCastException if key is not of type K 523 */ 524 @SuppressWarnings("unchecked") 525 @Override 526 public K get(Object key) { 527 return (K)key; 528 } 529 530 /** 531 * Always returns the empty set 532 */ 533 @Override 534 public Set<java.util.Map.Entry<K, K>> entrySet() { 535 return Collections.emptySet(); 536 } 537 538 @Override 539 public boolean containsKey(Object key) { 540 return true; 541 } 542 } 543 544 /** 545 * Fundamentally, an alignment is just a list of aligned residues in each 546 * protein. This method converts two lists of ResidueNumbers into an 547 * AFPChain. 548 * 549 * <p>Parameters are filled with defaults (often null) or sometimes 550 * calculated. 551 * 552 * <p>For a way to modify the alignment of an existing AFPChain, see 553 * {@link AlignmentTools#replaceOptAln(AFPChain, Atom[], Atom[], Map)} 554 * @param ca1 CA atoms of the first protein 555 * @param ca2 CA atoms of the second protein 556 * @param aligned1 A list of aligned residues from the first protein 557 * @param aligned2 A list of aligned residues from the second protein. 558 * Must be the same length as aligned1. 559 * @return An AFPChain representing the alignment. Many properties may be 560 * null or another default. 561 * @throws StructureException if an error occured during superposition 562 * @throws IllegalArgumentException if aligned1 and aligned2 have different 563 * lengths 564 * @see AlignmentTools#replaceOptAln(AFPChain, Atom[], Atom[], Map) 565 */ 566 public static AFPChain createAFPChain(Atom[] ca1, Atom[] ca2, 567 ResidueNumber[] aligned1, ResidueNumber[] aligned2 ) throws StructureException { 568 //input validation 569 int alnLen = aligned1.length; 570 if(alnLen != aligned2.length) { 571 throw new IllegalArgumentException("Alignment lengths are not equal"); 572 } 573 574 AFPChain a = new AFPChain(AFPChain.UNKNOWN_ALGORITHM); 575 try { 576 a.setName1(ca1[0].getGroup().getChain().getStructure().getName()); 577 if(ca2[0].getGroup().getChain().getStructure() != null) { 578 // common case for cloned ca2 579 a.setName2(ca2[0].getGroup().getChain().getStructure().getName()); 580 } 581 } catch(Exception e) { 582 // One of the structures wasn't fully created. Ignore 583 } 584 a.setBlockNum(1); 585 a.setCa1Length(ca1.length); 586 a.setCa2Length(ca2.length); 587 588 a.setOptLength(alnLen); 589 a.setOptLen(new int[] {alnLen}); 590 591 592 Matrix[] ms = new Matrix[a.getBlockNum()]; 593 a.setBlockRotationMatrix(ms); 594 Atom[] blockShiftVector = new Atom[a.getBlockNum()]; 595 a.setBlockShiftVector(blockShiftVector); 596 597 String[][][] pdbAln = new String[1][2][alnLen]; 598 for(int i=0;i<alnLen;i++) { 599 pdbAln[0][0][i] = aligned1[i].getChainName()+":"+aligned1[i]; 600 pdbAln[0][1][i] = aligned2[i].getChainName()+":"+aligned2[i]; 601 } 602 603 a.setPdbAln(pdbAln); 604 605 // convert pdbAln to optAln, and fill in some other basic parameters 606 AFPChainXMLParser.rebuildAFPChain(a, ca1, ca2); 607 608 return a; 609 610 // Currently a single block. Split into several blocks by sequence if needed 611 // return AlignmentTools.splitBlocksByTopology(a,ca1,ca2); 612 } 613 614 /** 615 * 616 * @param a 617 * @param ca1 618 * @param ca2 619 * @return 620 * @throws StructureException if an error occurred during superposition 621 */ 622 public static AFPChain splitBlocksByTopology(AFPChain a, Atom[] ca1, Atom[] ca2) throws StructureException { 623 int[][][] optAln = a.getOptAln(); 624 int blockNum = a.getBlockNum(); 625 int[] optLen = a.getOptLen(); 626 627 // Determine block lengths 628 // Split blocks if residue indices don't increase monotonically 629 List<Integer> newBlkLen = new ArrayList<>(); 630 boolean blockChanged = false; 631 for(int blk=0;blk<blockNum;blk++) { 632 int currLen=1; 633 for(int pos=1;pos<optLen[blk];pos++) { 634 if( optAln[blk][0][pos] <= optAln[blk][0][pos-1] 635 || optAln[blk][1][pos] <= optAln[blk][1][pos-1] ) 636 { 637 //start a new block 638 newBlkLen.add(currLen); 639 currLen = 0; 640 blockChanged = true; 641 } 642 currLen++; 643 } 644 if(optLen[blk] < 2 ) { 645 newBlkLen.add(optLen[blk]); 646 } else { 647 newBlkLen.add(currLen); 648 } 649 } 650 651 // Check if anything needs to be split 652 if( !blockChanged ) { 653 return a; 654 } 655 656 // Split blocks 657 List<int[][]> blocks = new ArrayList<>( newBlkLen.size() ); 658 659 int oldBlk = 0; 660 int pos = 0; 661 for(int blkLen : newBlkLen) { 662 if( blkLen == optLen[oldBlk] ) { 663 assert(pos == 0); //should be the whole block 664 // Use the old block 665 blocks.add(optAln[oldBlk]); 666 } else { 667 int[][] newBlock = new int[2][blkLen]; 668 assert( pos+blkLen <= optLen[oldBlk] ); // don't overrun block 669 for(int i=0; i<blkLen;i++) { 670 newBlock[0][i] = optAln[oldBlk][0][pos + i]; 671 newBlock[1][i] = optAln[oldBlk][1][pos + i]; 672 } 673 pos += blkLen; 674 blocks.add(newBlock); 675 676 if( pos == optLen[oldBlk] ) { 677 // Finished this oldBlk, start the next 678 oldBlk++; 679 pos = 0; 680 } 681 } 682 } 683 684 // Store new blocks 685 int[][][] newOptAln = blocks.toArray(new int[0][][]); 686 int[] newBlockLens = new int[newBlkLen.size()]; 687 for(int i=0;i<newBlkLen.size();i++) { 688 newBlockLens[i] = newBlkLen.get(i); 689 } 690 691 return replaceOptAln(a, ca1, ca2, blocks.size(), newBlockLens, newOptAln); 692 } 693 694 /** 695 * It replaces an optimal alignment of an AFPChain and calculates all the new alignment scores and variables. 696 */ 697 public static AFPChain replaceOptAln(int[][][] newAlgn, AFPChain afpChain, Atom[] ca1, Atom[] ca2) throws StructureException { 698 699 //The order is the number of groups in the newAlgn 700 int order = newAlgn.length; 701 702 //Calculate the alignment length from all the subunits lengths 703 int[] optLens = new int[order]; 704 for(int s=0;s<order;s++) { 705 optLens[s] = newAlgn[s][0].length; 706 } 707 int optLength = 0; 708 for(int s=0;s<order;s++) { 709 optLength += optLens[s]; 710 } 711 712 //Create a copy of the original AFPChain and set everything needed for the structure update 713 AFPChain copyAFP = (AFPChain) afpChain.clone(); 714 715 //Set the new parameters of the optimal alignment 716 copyAFP.setOptLength(optLength); 717 copyAFP.setOptLen(optLens); 718 copyAFP.setOptAln(newAlgn); 719 720 //Set the block information of the new alignment 721 copyAFP.setBlockNum(order); 722 copyAFP.setBlockSize(optLens); 723 copyAFP.setBlockResList(newAlgn); 724 copyAFP.setBlockResSize(optLens); 725 copyAFP.setBlockGap(calculateBlockGap(newAlgn)); 726 727 //Recalculate properties: superposition, tm-score, etc 728 Atom[] ca2clone = StructureTools.cloneAtomArray(ca2); // don't modify ca2 positions 729 AlignmentTools.updateSuperposition(copyAFP, ca1, ca2clone); 730 731 //It re-does the sequence alignment strings from the OptAlgn information only 732 copyAFP.setAlnsymb(null); 733 AFPAlignmentDisplay.getAlign(copyAFP, ca1, ca2clone); 734 735 return copyAFP; 736 } 737 738 /** 739 * Takes an AFPChain and replaces the optimal alignment based on an alignment map 740 * 741 * <p>Parameters are filled with defaults (often null) or sometimes 742 * calculated. 743 * 744 * <p>For a way to create a new AFPChain, see 745 * {@link AlignmentTools#createAFPChain(Atom[], Atom[], ResidueNumber[], ResidueNumber[])} 746 * 747 * @param afpChain The alignment to be modified 748 * @param alignment The new alignment, as a Map 749 * @throws StructureException if an error occurred during superposition 750 * @see AlignmentTools#createAFPChain(Atom[], Atom[], ResidueNumber[], ResidueNumber[]) 751 */ 752 public static AFPChain replaceOptAln(AFPChain afpChain, Atom[] ca1, Atom[] ca2, 753 Map<Integer, Integer> alignment) throws StructureException { 754 755 // Determine block lengths 756 // Sort ca1 indices, then start a new block whenever ca2 indices aren't 757 // increasing monotonically. 758 Integer[] res1 = alignment.keySet().toArray(new Integer[0]); 759 Arrays.sort(res1); 760 List<Integer> blockLens = new ArrayList<>(2); 761 int optLength = 0; 762 Integer lastRes = alignment.get(res1[0]); 763 int blkLen = lastRes==null?0:1; 764 for(int i=1;i<res1.length;i++) { 765 Integer currRes = alignment.get(res1[i]); //res2 index 766 assert(currRes != null);// could be converted to if statement if assertion doesn't hold; just modify below as well. 767 if(lastRes<currRes) { 768 blkLen++; 769 } else { 770 // CP! 771 blockLens.add(blkLen); 772 optLength+=blkLen; 773 blkLen = 1; 774 } 775 lastRes = currRes; 776 } 777 blockLens.add(blkLen); 778 optLength+=blkLen; 779 780 // Create array structure for alignment 781 int[][][] optAln = new int[blockLens.size()][][]; 782 int pos1 = 0; //index into res1 783 for(int blk=0;blk<blockLens.size();blk++) { 784 optAln[blk] = new int[2][]; 785 blkLen = blockLens.get(blk); 786 optAln[blk][0] = new int[blkLen]; 787 optAln[blk][1] = new int[blkLen]; 788 int pos = 0; //index into optAln 789 while(pos<blkLen) { 790 optAln[blk][0][pos]=res1[pos1]; 791 Integer currRes = alignment.get(res1[pos1]); 792 optAln[blk][1][pos]=currRes; 793 pos++; 794 pos1++; 795 } 796 } 797 assert(pos1 == optLength); 798 799 // Create length array 800 int[] optLens = new int[blockLens.size()]; 801 for(int i=0;i<blockLens.size();i++) { 802 optLens[i] = blockLens.get(i); 803 } 804 805 return replaceOptAln(afpChain, ca1, ca2, blockLens.size(), optLens, optAln); 806 } 807 808 /** 809 * @param afpChain Input afpchain. UNMODIFIED 810 * @param ca1 811 * @param ca2 812 * @param optLens 813 * @param optAln 814 * @return A NEW AfpChain based off the input but with the optAln modified 815 * @throws StructureException if an error occured during superposition 816 */ 817 public static AFPChain replaceOptAln(AFPChain afpChain, Atom[] ca1, Atom[] ca2, 818 int blockNum, int[] optLens, int[][][] optAln) throws StructureException { 819 int optLength = 0; 820 for( int blk=0;blk<blockNum;blk++) { 821 optLength += optLens[blk]; 822 } 823 824 //set everything 825 AFPChain refinedAFP = (AFPChain) afpChain.clone(); 826 refinedAFP.setOptLength(optLength); 827 refinedAFP.setBlockSize(optLens); 828 refinedAFP.setOptLen(optLens); 829 refinedAFP.setOptAln(optAln); 830 refinedAFP.setBlockNum(blockNum); 831 832 //TODO recalculate properties: superposition, tm-score, etc 833 Atom[] ca2clone = StructureTools.cloneAtomArray(ca2); // don't modify ca2 positions 834 AlignmentTools.updateSuperposition(refinedAFP, ca1, ca2clone); 835 836 AFPAlignmentDisplay.getAlign(refinedAFP, ca1, ca2clone); 837 return refinedAFP; 838 } 839 840 841 /** 842 * After the alignment changes (optAln, optLen, blockNum, at a minimum), 843 * many other properties which depend on the superposition will be invalid. 844 * 845 * This method re-runs a rigid superposition over the whole alignment 846 * and repopulates the required properties, including RMSD (TotalRMSD) and 847 * TM-Score. 848 * @param afpChain 849 * @param ca1 850 * @param ca2 Second set of ca atoms. Will be modified based on the superposition 851 * @throws StructureException 852 * @see CECalculator#calc_rmsd(Atom[], Atom[], int, boolean) 853 * contains much of the same code, but stores results in a CECalculator 854 * instance rather than an AFPChain 855 */ 856 public static void updateSuperposition(AFPChain afpChain, Atom[] ca1, 857 Atom[] ca2) throws StructureException { 858 859 //Update ca information, because the atom array might also be changed 860 afpChain.setCa1Length(ca1.length); 861 afpChain.setCa2Length(ca2.length); 862 863 //We need this to get the correct superposition 864 int[] focusRes1 = afpChain.getFocusRes1(); 865 int[] focusRes2 = afpChain.getFocusRes2(); 866 if (focusRes1 == null) { 867 focusRes1 = new int[afpChain.getCa1Length()]; 868 afpChain.setFocusRes1(focusRes1); 869 } 870 if (focusRes2 == null) { 871 focusRes2 = new int[afpChain.getCa2Length()]; 872 afpChain.setFocusRes2(focusRes2); 873 } 874 875 if (afpChain.getNrEQR() == 0) return; 876 877 // create new arrays for the subset of atoms in the alignment. 878 Atom[] ca1aligned = new Atom[afpChain.getOptLength()]; 879 Atom[] ca2aligned = new Atom[afpChain.getOptLength()]; 880 881 fillAlignedAtomArrays(afpChain, ca1, ca2, ca1aligned, ca2aligned); 882 883 //Superimpose the two structures in correspondance to the new alignment 884 Matrix4d trans = SuperPositions.superpose(Calc.atomsToPoints(ca1aligned), 885 Calc.atomsToPoints(ca2aligned)); 886 887 Matrix matrix = Matrices.getRotationJAMA(trans); 888 Atom shift = Calc.getTranslationVector(trans); 889 890 Matrix[] blockMxs = new Matrix[afpChain.getBlockNum()]; 891 Arrays.fill(blockMxs, matrix); 892 afpChain.setBlockRotationMatrix(blockMxs); 893 Atom[] blockShifts = new Atom[afpChain.getBlockNum()]; 894 Arrays.fill(blockShifts, shift); 895 afpChain.setBlockShiftVector(blockShifts); 896 897 for (Atom a : ca2aligned) { 898 Calc.rotate(a, matrix); 899 Calc.shift(a, shift); 900 } 901 902 //Calculate the RMSD and TM score for the new alignment 903 double rmsd = Calc.rmsd(ca1aligned, ca2aligned); 904 double tmScore = Calc.getTMScore(ca1aligned, ca2aligned, ca1.length, ca2.length); 905 afpChain.setTotalRmsdOpt(rmsd); 906 afpChain.setTMScore(tmScore); 907 908 int[] blockLens = afpChain.getOptLen(); 909 int[][][] optAln = afpChain.getOptAln(); 910 911 //Calculate the RMSD and TM score for every block of the new alignment 912 double[] blockRMSD = new double[afpChain.getBlockNum()]; 913 double[] blockScore = new double[afpChain.getBlockNum()]; 914 for (int k=0; k<afpChain.getBlockNum(); k++){ 915 //Create the atom arrays corresponding to the aligned residues in the block 916 Atom[] ca1block = new Atom[afpChain.getOptLen()[k]]; 917 Atom[] ca2block = new Atom[afpChain.getOptLen()[k]]; 918 int position=0; 919 for(int i=0;i<blockLens[k];i++) { 920 int pos1 = optAln[k][0][i]; 921 int pos2 = optAln[k][1][i]; 922 Atom a1 = ca1[pos1]; 923 Atom a2 = (Atom) ca2[pos2].clone(); 924 ca1block[position] = a1; 925 ca2block[position] = a2; 926 position++; 927 } 928 if (position != afpChain.getOptLen()[k]){ 929 logger.warn("AFPChainScorer getTMScore: Problems reconstructing block alignment! nr of loaded atoms is " + position + " but should be " + afpChain.getOptLen()[k]); 930 // we need to resize the array, because we allocated too many atoms earlier on. 931 ca1block = (Atom[]) resizeArray(ca1block, position); 932 ca2block = (Atom[]) resizeArray(ca2block, position); 933 } 934 //Superimpose the two block structures 935 Matrix4d transb = SuperPositions.superpose(Calc.atomsToPoints(ca1block), 936 Calc.atomsToPoints(ca2block)); 937 938 blockMxs[k] = Matrices.getRotationJAMA(trans); 939 blockShifts[k] = Calc.getTranslationVector(trans); 940 941 Calc.transform(ca2block, transb); 942 943 //Calculate the RMSD and TM score for the block 944 double rmsdb = Calc.rmsd(ca1block, ca2block); 945 double tmScoreb = Calc.getTMScore(ca1block, ca2block, ca1.length, ca2.length); 946 blockRMSD[k] = rmsdb; 947 blockScore[k] = tmScoreb; 948 } 949 afpChain.setOptRmsd(blockRMSD); 950 afpChain.setBlockRmsd(blockRMSD); 951 afpChain.setBlockScore(blockScore); 952 } 953 954 /** 955 * Reallocates an array with a new size, and copies the contents 956 * of the old array to the new array. 957 * @param oldArray the old array, to be reallocated. 958 * @param newSize the new array size. 959 * @return A new array with the same contents. 960 */ 961 public static Object resizeArray (Object oldArray, int newSize) { 962 int oldSize = java.lang.reflect.Array.getLength(oldArray); 963 @SuppressWarnings("rawtypes") 964 Class elementType = oldArray.getClass().getComponentType(); 965 Object newArray = java.lang.reflect.Array.newInstance( 966 elementType,newSize); 967 int preserveLength = Math.min(oldSize,newSize); 968 if (preserveLength > 0) 969 System.arraycopy (oldArray,0,newArray,0,preserveLength); 970 return newArray; 971 } 972 973 /** 974 * Print an alignment map in a concise representation. Edges are given 975 * as two numbers separated by '>'. They are chained together where possible, 976 * or separated by spaces where disjoint or branched. 977 * 978 * <p>Note that more concise representations may be possible.</p> 979 * 980 * Examples: 981 * <ul> 982 * <li>1>2>3>1</li> 983 * <li>1>2>3>2 4>3</li> 984 * </ul> 985 * @param alignment The input function, as a map (see {@link AlignmentTools#alignmentAsMap(AFPChain)}) 986 * @param identity An identity-like function providing the isomorphism between 987 * the codomain of alignment (of type T) and the domain (type S). 988 * @return 989 */ 990 public static <S,T> String toConciseAlignmentString(Map<S,T> alignment, Map<T,S> identity) { 991 // Clone input to prevent changes 992 Map<S,T> alig = new HashMap<>(alignment); 993 994 // Generate inverse alignment 995 Map<S,List<S>> inverse = new HashMap<>(); 996 for(Entry<S,T> e: alig.entrySet()) { 997 S val = identity.get(e.getValue()); 998 if( inverse.containsKey(val) ) { 999 List<S> l = inverse.get(val); 1000 l.add(e.getKey()); 1001 } else { 1002 List<S> l = new ArrayList<>(); 1003 l.add(e.getKey()); 1004 inverse.put(val,l); 1005 } 1006 } 1007 1008 StringBuilder str = new StringBuilder(); 1009 1010 while(!alig.isEmpty()){ 1011 // Pick an edge and work upstream to a root or cycle 1012 S seedNode = alig.keySet().iterator().next(); 1013 S node = seedNode; 1014 if( inverse.containsKey(seedNode)) { 1015 node = inverse.get(seedNode).iterator().next(); 1016 while( node != seedNode && inverse.containsKey(node)) { 1017 node = inverse.get(node).iterator().next(); 1018 } 1019 } 1020 1021 // Now work downstream, deleting edges as we go 1022 seedNode = node; 1023 str.append(node); 1024 1025 while(alig.containsKey(node)) { 1026 S lastNode = node; 1027 node = identity.get( alig.get(lastNode) ); 1028 1029 // Output 1030 str.append('>'); 1031 str.append(node); 1032 1033 // Remove edge 1034 alig.remove(lastNode); 1035 List<S> inv = inverse.get(node); 1036 if(inv.size() > 1) { 1037 inv.remove(node); 1038 } else { 1039 inverse.remove(node); 1040 } 1041 } 1042 if(!alig.isEmpty()) { 1043 str.append(' '); 1044 } 1045 } 1046 1047 return str.toString(); 1048 } 1049 1050 /** 1051 * @see #toConciseAlignmentString(Map, Map) 1052 */ 1053 public static <T> String toConciseAlignmentString(Map<T, T> alignment) { 1054 return toConciseAlignmentString(alignment, new IdentityMap<T>()); 1055 } 1056 1057 /** 1058 * @see #toConciseAlignmentString(Map, Map) 1059 */ 1060 public static Map<Integer, Integer> fromConciseAlignmentString(String string) { 1061 Map<Integer, Integer> map = new HashMap<>(); 1062 boolean matches = true; 1063 while (matches) { 1064 Pattern pattern = Pattern.compile("(\\d+)>(\\d+)"); 1065 Matcher matcher = pattern.matcher(string); 1066 matches = matcher.find(); 1067 if (matches) { 1068 Integer from = Integer.parseInt(matcher.group(1)); 1069 Integer to = Integer.parseInt(matcher.group(2)); 1070 map.put(from, to); 1071 string = string.substring(matcher.end(1) + 1); 1072 } 1073 } 1074 return map; 1075 } 1076 1077 /** 1078 * Method that calculates the number of gaps in each subunit block of an optimal AFP alignment. 1079 * @param optAln 1080 * an optimal alignment in the format int[][][] 1081 * @return an int[] array of order length containing the gaps in each block as int[block] 1082 */ 1083 public static int[] calculateBlockGap(int[][][] optAln){ 1084 1085 //Initialize the array to be returned 1086 int [] blockGap = new int[optAln.length]; 1087 1088 //Loop for every block and look in both chains for non-contiguous residues. 1089 for (int i=0; i<optAln.length; i++){ 1090 int gaps = 0; //the number of gaps in that block 1091 int last1 = 0; //the last residue position in chain 1 1092 int last2 = 0; //the last residue position in chain 2 1093 //Loop for every position in the block 1094 for (int j=0; j<optAln[i][0].length; j++){ 1095 //If the first position is evaluated initialize the last positions 1096 if (j==0){ 1097 last1 = optAln[i][0][j]; 1098 last2 = optAln[i][1][j]; 1099 } 1100 else{ 1101 //If one of the positions or both are not contiguous increment the number of gaps 1102 if (optAln[i][0][j] > last1+1 || optAln[i][1][j] > last2+1){ 1103 gaps++; 1104 last1 = optAln[i][0][j]; 1105 last2 = optAln[i][1][j]; 1106 } 1107 //Otherwise just set the last position to the current one 1108 else{ 1109 last1 = optAln[i][0][j]; 1110 last2 = optAln[i][1][j]; 1111 } 1112 } 1113 } 1114 blockGap[i] = gaps; 1115 } 1116 return blockGap; 1117 } 1118 1119 /** 1120 * Creates a simple interaction format (SIF) file for an alignment. 1121 * 1122 * The SIF file can be read by network software (eg Cytoscape) to analyze 1123 * alignments as graphs. 1124 * 1125 * This function creates a graph with residues as nodes and two types of edges: 1126 * 1. backbone edges, which connect adjacent residues in the aligned protein 1127 * 2. alignment edges, which connect aligned residues 1128 * 1129 * @param out Stream to write to 1130 * @param afpChain alignment to write 1131 * @param ca1 First protein, used to generate node names 1132 * @param ca2 Second protein, used to generate node names 1133 * @param backboneInteraction Two-letter string used to identify backbone edges 1134 * @param alignmentInteraction Two-letter string used to identify alignment edges 1135 * @throws IOException 1136 */ 1137 public static void alignmentToSIF(Writer out,AFPChain afpChain, 1138 Atom[] ca1,Atom[] ca2, String backboneInteraction, 1139 String alignmentInteraction) throws IOException { 1140 1141 //out.write("Res1\tInteraction\tRes2\n"); 1142 String name1 = afpChain.getName1(); 1143 String name2 = afpChain.getName2(); 1144 if(name1==null) name1=""; else name1+=":"; 1145 if(name2==null) name2=""; else name2+=":"; 1146 1147 // Print alignment edges 1148 int nblocks = afpChain.getBlockNum(); 1149 int[] blockLen = afpChain.getOptLen(); 1150 int[][][] optAlign = afpChain.getOptAln(); 1151 for(int b=0;b<nblocks;b++) { 1152 for(int r=0;r<blockLen[b];r++) { 1153 int res1 = optAlign[b][0][r]; 1154 int res2 = optAlign[b][1][r]; 1155 1156 ResidueNumber rn1 = ca1[res1].getGroup().getResidueNumber(); 1157 ResidueNumber rn2 = ca2[res2].getGroup().getResidueNumber(); 1158 1159 String node1 = name1+rn1.getChainName()+rn1.toString(); 1160 String node2 = name2+rn2.getChainName()+rn2.toString(); 1161 1162 out.write(String.format("%s\t%s\t%s\n",node1, alignmentInteraction, node2)); 1163 } 1164 } 1165 1166 // Print first backbone edges 1167 ResidueNumber rn = ca1[0].getGroup().getResidueNumber(); 1168 String last = name1+rn.getChainName()+rn.toString(); 1169 for(int i=1;i<ca1.length;i++) { 1170 rn = ca1[i].getGroup().getResidueNumber(); 1171 String curr = name1+rn.getChainName()+rn.toString(); 1172 out.write(String.format("%s\t%s\t%s\n",last, backboneInteraction, curr)); 1173 last = curr; 1174 } 1175 1176 // Print second backbone edges, if the proteins differ 1177 // Do some quick checks for whether the proteins differ 1178 // (Not perfect, but should detect major differences and CPs.) 1179 if(!name1.equals(name2) || 1180 ca1.length!=ca2.length || 1181 (ca1.length>0 && ca1[0].getGroup()!=null && ca2[0].getGroup()!=null && 1182 !ca1[0].getGroup().getResidueNumber().equals(ca2[0].getGroup().getResidueNumber()) ) ) { 1183 rn = ca2[0].getGroup().getResidueNumber(); 1184 last = name2+rn.getChainName()+rn.toString(); 1185 for(int i=1;i<ca2.length;i++) { 1186 rn = ca2[i].getGroup().getResidueNumber(); 1187 String curr = name2+rn.getChainName()+rn.toString(); 1188 out.write(String.format("%s\t%s\t%s\n",last, backboneInteraction, curr)); 1189 last = curr; 1190 } 1191 } 1192 } 1193 1194 1195 1196 /** get an artificial List of chains containing the Atoms and groups. 1197 * Does NOT rotate anything. 1198 * @param ca 1199 * @return a list of Chains that is built up from the Atoms in the ca array 1200 */ 1201 public static final List<Chain> getAlignedModel(Atom[] ca){ 1202 1203 List<Chain> model = new ArrayList<>(); 1204 for ( Atom a: ca){ 1205 1206 Group g = a.getGroup(); 1207 Chain parentC = g.getChain(); 1208 1209 Chain newChain = null; 1210 for ( Chain c : model) { 1211 if ( c.getId().equals(parentC.getId())){ 1212 newChain = c; 1213 break; 1214 } 1215 } 1216 if ( newChain == null){ 1217 1218 newChain = new ChainImpl(); 1219 1220 newChain.setId(parentC.getId()); 1221 1222 model.add(newChain); 1223 } 1224 1225 newChain.addGroup(g); 1226 1227 } 1228 1229 return model; 1230 } 1231 1232 1233 /** Get an artifical Structure containing both chains. 1234 * Does NOT rotate anything 1235 * @param ca1 1236 * @param ca2 1237 * @return a structure object containing two models, one for each set of Atoms. 1238 * @throws StructureException 1239 */ 1240 public static final Structure getAlignedStructure(Atom[] ca1, Atom[] ca2) throws StructureException{ 1241 1242 /* Previous implementation commented 1243 1244 Structure s = new StructureImpl(); 1245 1246 1247 List<Chain>model1 = getAlignedModel(ca1); 1248 List<Chain>model2 = getAlignedModel(ca2); 1249 s.addModel(model1); 1250 s.addModel(model2); 1251 1252 return s;*/ 1253 1254 Structure s = new StructureImpl(); 1255 1256 List<Chain>model1 = getAlignedModel(ca1); 1257 s.addModel(model1); 1258 List<Chain> model2 = getAlignedModel(ca2); 1259 s.addModel(model2); 1260 1261 return s; 1262 } 1263 1264 /** Rotate the Atoms/Groups so they are aligned for the 3D visualisation 1265 * 1266 * @param afpChain 1267 * @param ca1 1268 * @param ca2 1269 * @return an array of Groups that are transformed for 3D display 1270 * @throws StructureException 1271 */ 1272 public static Group[] prepareGroupsForDisplay(AFPChain afpChain, Atom[] ca1, Atom[] ca2) throws StructureException{ 1273 1274 1275 if ( afpChain.getBlockRotationMatrix().length == 0 ) { 1276 // probably the alignment is too short! 1277 System.err.println("No rotation matrix found to rotate 2nd structure!"); 1278 afpChain.setBlockRotationMatrix(new Matrix[]{Matrix.identity(3, 3)}); 1279 afpChain.setBlockShiftVector(new Atom[]{new AtomImpl()}); 1280 } 1281 1282 // List of groups to be rotated according to the alignment 1283 Group[] twistedGroups = new Group[ ca2.length]; 1284 1285 //int blockNum = afpChain.getBlockNum(); 1286 1287 int i = -1; 1288 1289 // List of groups from the structure not included in ca2 (e.g. ligands) 1290 // Will be rotated according to first block 1291 List<Group> hetatms2 = StructureTools.getUnalignedGroups(ca2); 1292 1293 if ( (afpChain.getAlgorithmName().equals(FatCatRigid.algorithmName) ) || (afpChain.getAlgorithmName().equals(FatCatFlexible.algorithmName) ) ){ 1294 1295 for (Atom a: ca2){ 1296 i++; 1297 twistedGroups[i]=a.getGroup(); 1298 1299 } 1300 1301 twistedGroups = AFPTwister.twistOptimized(afpChain, ca1, ca2); 1302 1303 //} else if (( blockNum == 1 ) || (afpChain.getAlgorithmName().equals(CeCPMain.algorithmName))) { 1304 } else { 1305 1306 Matrix m = afpChain.getBlockRotationMatrix()[ 0]; 1307 Atom shift = afpChain.getBlockShiftVector() [ 0 ]; 1308 1309 shiftCA2(afpChain, ca2, m,shift, twistedGroups); 1310 1311 } 1312 1313 if ( afpChain.getBlockNum() > 0){ 1314 1315 // Superimpose ligands relative to the first block 1316 if( hetatms2.size() > 0 ) { 1317 1318 if ( afpChain.getBlockRotationMatrix().length > 0 ) { 1319 1320 Matrix m1 = afpChain.getBlockRotationMatrix()[0]; 1321 //m1.print(3,3); 1322 Atom vector1 = afpChain.getBlockShiftVector()[0]; 1323 //System.out.println("shift vector:" + vector1); 1324 1325 for ( Group g : hetatms2){ 1326 Calc.rotate(g, m1); 1327 Calc.shift(g,vector1); 1328 } 1329 } 1330 } 1331 } 1332 1333 return twistedGroups; 1334 } 1335 1336 /** only shift CA positions. 1337 * 1338 */ 1339 public static void shiftCA2(AFPChain afpChain, Atom[] ca2, Matrix m, Atom shift, Group[] twistedGroups) { 1340 1341 int i = -1; 1342 for (Atom a: ca2){ 1343 i++; 1344 Group g = a.getGroup(); 1345 1346 Calc.rotate(g,m); 1347 Calc.shift(g, shift); 1348 1349 if (g.hasAltLoc()){ 1350 for (Group alt: g.getAltLocs()){ 1351 for (Atom alta : alt.getAtoms()){ 1352 if ( g.getAtoms().contains(alta)) 1353 continue; 1354 Calc.rotate(alta,m); 1355 Calc.shift(alta,shift); 1356 } 1357 } 1358 } 1359 twistedGroups[i]=g; 1360 } 1361 } 1362 1363 /** 1364 * Fill the aligned Atom arrays with the equivalent residues in the afpChain. 1365 * @param afpChain 1366 * @param ca1 1367 * @param ca2 1368 * @param ca1aligned 1369 * @param ca2aligned 1370 */ 1371 public static void fillAlignedAtomArrays(AFPChain afpChain, Atom[] ca1, 1372 Atom[] ca2, Atom[] ca1aligned, Atom[] ca2aligned) { 1373 1374 int pos=0; 1375 int[] blockLens = afpChain.getOptLen(); 1376 int[][][] optAln = afpChain.getOptAln(); 1377 assert(afpChain.getBlockNum() <= optAln.length); 1378 1379 for (int block=0; block < afpChain.getBlockNum(); block++) { 1380 for(int i=0;i<blockLens[block];i++) { 1381 int pos1 = optAln[block][0][i]; 1382 int pos2 = optAln[block][1][i]; 1383 Atom a1 = ca1[pos1]; 1384 Atom a2 = (Atom) ca2[pos2].clone(); 1385 ca1aligned[pos] = a1; 1386 ca2aligned[pos] = a2; 1387 pos++; 1388 } 1389 } 1390 1391 // this can happen when we load an old XML serialization which did not support modern ChemComp representation of modified residues. 1392 if (pos != afpChain.getOptLength()){ 1393 logger.warn("AFPChainScorer getTMScore: Problems reconstructing alignment! nr of loaded atoms is " + pos + " but should be " + afpChain.getOptLength()); 1394 // we need to resize the array, because we allocated too many atoms earlier on. 1395 ca1aligned = (Atom[]) resizeArray(ca1aligned, pos); 1396 ca2aligned = (Atom[]) resizeArray(ca2aligned, pos); 1397 } 1398 1399 } 1400 1401 /** 1402 * Find the alignment position with the highest atomic distance between the 1403 * equivalent atomic positions of the arrays and remove it from the 1404 * alignment. 1405 * 1406 * @param afpChain 1407 * original alignment, will be modified 1408 * @param ca1 1409 * atom array, will not be modified 1410 * @param ca2 1411 * atom array, will not be modified 1412 * @return the original alignment, with the alignment position at the 1413 * highest distance removed 1414 * @throws StructureException 1415 */ 1416 public static AFPChain deleteHighestDistanceColumn(AFPChain afpChain, 1417 Atom[] ca1, Atom[] ca2) throws StructureException { 1418 1419 int[][][] optAln = afpChain.getOptAln(); 1420 1421 int maxBlock = 0; 1422 int maxPos = 0; 1423 double maxDistance = Double.MIN_VALUE; 1424 1425 for (int b = 0; b < optAln.length; b++) { 1426 for (int p = 0; p < optAln[b][0].length; p++) { 1427 Atom ca2clone = ca2[optAln[b][1][p]]; 1428 Calc.rotate(ca2clone, afpChain.getBlockRotationMatrix()[b]); 1429 Calc.shift(ca2clone, afpChain.getBlockShiftVector()[b]); 1430 1431 double distance = Calc.getDistance(ca1[optAln[b][0][p]], 1432 ca2clone); 1433 if (distance > maxDistance) { 1434 maxBlock = b; 1435 maxPos = p; 1436 maxDistance = distance; 1437 } 1438 } 1439 } 1440 1441 return deleteColumn(afpChain, ca1, ca2, maxBlock, maxPos); 1442 } 1443 1444 /** 1445 * Delete an alignment position from the original alignment object. 1446 * 1447 * @param afpChain 1448 * original alignment, will be modified 1449 * @param ca1 1450 * atom array, will not be modified 1451 * @param ca2 1452 * atom array, will not be modified 1453 * @param block 1454 * block of the alignment position 1455 * @param pos 1456 * position index in the block 1457 * @return the original alignment, with the alignment position removed 1458 * @throws StructureException 1459 */ 1460 public static AFPChain deleteColumn(AFPChain afpChain, Atom[] ca1, 1461 Atom[] ca2, int block, int pos) throws StructureException { 1462 1463 // Check validity of the inputs 1464 if (afpChain.getBlockNum() <= block) { 1465 throw new IndexOutOfBoundsException(String.format( 1466 "Block index requested (%d) is higher than the total number of AFPChain blocks (%d).", 1467 block, afpChain.getBlockNum())); 1468 } 1469 if (afpChain.getOptAln()[block][0].length <= pos) { 1470 throw new IndexOutOfBoundsException(String.format( 1471 "Position index requested (%d) is higher than the total number of aligned position in the AFPChain block (%d).", 1472 block, afpChain.getBlockSize()[block])); 1473 } 1474 1475 int[][][] optAln = afpChain.getOptAln(); 1476 1477 int[] newPos0 = new int[optAln[block][0].length - 1]; 1478 int[] newPos1 = new int[optAln[block][1].length - 1]; 1479 1480 int position = 0; 1481 for (int p = 0; p < optAln[block][0].length; p++) { 1482 1483 if (p == pos) 1484 continue; 1485 1486 newPos0[position] = optAln[block][0][p]; 1487 newPos1[position] = optAln[block][1][p]; 1488 1489 position++; 1490 } 1491 1492 optAln[block][0] = newPos0; 1493 optAln[block][1] = newPos1; 1494 1495 return AlignmentTools.replaceOptAln(optAln, afpChain, ca1, ca2); 1496 } 1497}