kmerstore.h 16 KB
Newer Older
Mikaël Salson's avatar
Mikaël Salson committed
1 2 3 4 5 6 7 8
#ifndef KMERSTORE_H
#define KMERSTORE_H

#include <map>
#include <string>
#include <list>
#include <stdexcept>
#include <stdint.h>
9
#include <math.h>
10
#include "bioreader.hpp"
Mikaël Salson's avatar
Mikaël Salson committed
11 12 13 14
#include "tools.h"

using namespace std;

15 16 17
typedef
enum { KMER_INDEX, AC_AUTOMATON } IndexTypes;

Mikaël Salson's avatar
Mikaël Salson committed
18 19 20 21 22
class Kmer {
public:
  unsigned int count;

  Kmer();
23 24 25 26

  /**
   * This constructor is used via a IKmerStore<Kmer> index (hence the argument list)       
   */
27
  Kmer(const string &label, int strand=1, size_t length=0);
28

Mikaël Salson's avatar
Mikaël Salson committed
29
  Kmer &operator+=(const Kmer &);
30

31 32 33 34 35
  /**
   * @return the length of the kmer (only there for compatibility with KmerAffect)
   */
  size_t getLength() const;

36 37 38 39
  /**
   * When indexing revcomp, should the value be the same or not?
   * In other words, does the information stored by the class is strand-dependent?
   */
Mikaël Salson's avatar
Mikaël Salson committed
40
  static bool hasRevcompSymetry();
41 42 43 44

  /**
   * @return true if the element is the same as when initialised with default constructor.
   */
45
  bool isNull() const;
46 47
	
	string getLabel() const;
48 49 50 51 52 53

  /**
   * @return true iff the kmer is unknown (which doesn't make sense here but
   * it only there for a reason of compatibility with KmerAffect)
   */
  bool isUnknown() const;
Mikaël Salson's avatar
Mikaël Salson committed
54 55
} ;
ostream &operator<<(ostream &os, const Kmer &kmer);
56 57 58 59 60 61 62
bool operator==(const Kmer &k1, const Kmer &k2);
bool operator<(const Kmer &k1, const Kmer &k2);
bool operator>(const Kmer &k1, const Kmer &k2);
bool operator<=(const Kmer &k1, const Kmer &k2);
bool operator>=(const Kmer &k1, const Kmer &k2);
bool operator!=(const Kmer &k1, const Kmer &k2);

Mikaël Salson's avatar
Mikaël Salson committed
63 64 65 66 67 68 69 70 71 72 73


/* K-mer indexing */

template <class T> class IKmerStore
{
protected:
  bool revcomp_indexed;
  int k; // weight of the seed
  int s; // span of the seed (s >= k)
  string seed ;
74
  size_t nb_kmers_inserted;
75
  size_t max_size_indexing;
76
  bool finished_building;
Mikaël Salson's avatar
Mikaël Salson committed
77 78 79 80 81

public:

  virtual ~IKmerStore();

82 83
  static int last_id;
  int id; // id of this index
84
  int refs; // number of germlines using this index
85

86
  list< pair <T, BioReader> > labels;
87

88 89
  IKmerStore();

Mikaël Salson's avatar
Mikaël Salson committed
90 91 92
  /**
   * @param input: A single FASTA file
   * @param label: label that must be associated to the given files
93
   * @param seed: the seed to use for indexing. By default it will be the seed of the index
94
   * @post All the sequences in the FASTA files have been indexed, and the label is stored in the list of labels
Mikaël Salson's avatar
Mikaël Salson committed
95
   */
96
  void insert(BioReader& input, const string& label="", int keep_only = 0, string seed = "");
Mikaël Salson's avatar
Mikaël Salson committed
97 98 99 100

  /**
   * @param input: A list of FASTA files
   * @param label: label that must be associated to the given files
101
   * @param seed: the seed to use for indexing. By default it will be the seed of the index
102
   * @post All the sequences in the FASTA files have been indexed, and the label is stored in the list of labels
Mikaël Salson's avatar
Mikaël Salson committed
103
   */
104
  void insert(list<BioReader>& input, const string& label="", int keep_only = 0, string seed = "");
Mikaël Salson's avatar
Mikaël Salson committed
105 106 107 108
  
  /**
   * @param input: A sequence to be cut in k-mers
   * @param label: label that must be associated to the given files
109 110 111 112
   * @param keep_only: if > 0 will keep at most the last keep_only nucleotides
   *                   of the sequence. if < 0 will keep at most the first
   *                   keep_only nucleotides of the sequence. if == 0,
   *                   will keep all the sequence.
113
   * @param seed: the seed to use for indexing. By default it will be the seed of the index. Also note that when the index ! hasDifferentKmerTypes(), the seed of the index will always be used (and therefore the parameter will be ignored)
Mikaël Salson's avatar
Mikaël Salson committed
114 115
   * @post All the k-mers in the sequence have been indexed.
   */
116 117 118 119 120 121 122 123 124 125
  virtual void insert(const seqtype &sequence,
                      const string &label,
                      bool ignore_extended_nucleotides=true, int keep_only = 0,
                      string seed="");

  /**
   * Perform extra steps to finish the building of the index. After using
   * that function, we are not supposed to insert any other sequence.
   * No query function should be called before calling that function.
   */
126
  virtual void finish_building();
Mikaël Salson's avatar
Mikaël Salson committed
127 128 129 130 131

  /**
   * @param word: a k-mer
   * @return the value for the kmer as stored in the structure (don't do revcomp)
   */ 
132
  virtual T& get(seqtype &word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
133

134
  /**
135 136 137 138
   * @return the percentage of kmers that are set in the index.
   *         When ! hasDifferentKmerTypes(), the index load is always the same
   *         (apart for the unknown kmer).
   *         When kmer.isUnknown(), we return the load of all th other kmers.
139
   */
140
  virtual float getIndexLoad(T kmer) const;
141

142 143 144 145 146 147
  /**
   * @return the given integer (size of a read),
   *         but limit this size to max_size_indexing if it was defined
   */
  int atMostMaxSizeIndexing(int n) const;

148 149 150
  /**
   * @return probability that the number of kmers is 'at_least' or more in a sequence of length 'length'
   */
151
  double getProbabilityAtLeastOrAbove(T kmer, int at_least, int length) const;
152

Mikaël Salson's avatar
Mikaël Salson committed
153 154 155 156 157 158 159 160 161 162
  /**
   * @return the value of k
   */
  int getK() const;

  /**
   * @return the value of s
   */
  int getS() const;

163 164 165 166 167
  /**
   * @return the seed used
   */
  string getSeed() const;

168 169 170 171
  /**
   * @param kmer: a kmer
   * @return one label associated with the kmer
   */
172
  BioReader getLabel(T kmer) const;
173

174 175 176 177 178
  /**
   * @return whether the index differentiate kmer types
   */
  virtual bool hasDifferentKmerTypes() const;

Mikaël Salson's avatar
Mikaël Salson committed
179 180 181 182
  /**
   * @param seq: a sequence
   * @param no_revcomp: force not to revcomp the sequence, even if
   *                    the index was built with revcomp.
183 184
   * @return a vector of length seq.length() - d, where d is the span
   *         of the seed (minus one) or is 0 when it doesn't apply.
185
   * containing for each position the corresponding value in the index.
Mikaël Salson's avatar
Mikaël Salson committed
186
   */
187
  virtual vector<T> getResults(const seqtype &seq, bool no_revcomp=false, string seed="") = 0;
Mikaël Salson's avatar
Mikaël Salson committed
188 189 190 191 192 193 194 195 196 197 198 199

  /**
   * @return true iff the revcomp is indexed
   */
  bool isRevcomp() const;

  /**
   * @param string: a k-mer
   * @return the value in the index associated to the given k-mer.
   *         isRevcomp() ==> the revcomp of the k-mer may be considered
   * @pre word.length() == this->k
   */
200
  virtual T& operator[](seqtype& word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
201 202
};

203 204 205
template<class T>
IKmerStore<T>::IKmerStore() {
  id = ++last_id;
Mikaël Salson's avatar
Mikaël Salson committed
206
  refs = 0;
207
  finished_building = false;
208 209
}

210
template<class T> int IKmerStore<T>::last_id = 0;
Mikaël Salson's avatar
Mikaël Salson committed
211 212 213 214 215

template <class T> 
class MapKmerStore : public IKmerStore<T>
{
public:	
216
  map<seqtype, T> store;
Mikaël Salson's avatar
Mikaël Salson committed
217 218 219 220 221 222 223

  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  MapKmerStore(string seed, bool=false);
  MapKmerStore(int k, bool=false);
224

225
  vector<T> getResults(const seqtype &seq, bool no_revcomp=false, string seed="");
226 227
  T& get(seqtype &word);
  T& operator[](seqtype & word);
228 229 230

 private:
  void init();
Mikaël Salson's avatar
Mikaël Salson committed
231 232 233 234 235 236 237
};

template <class T> 
class ArrayKmerStore : public IKmerStore<T>
{
  T* store;
	
238
  int index(const seqtype& word) const;
239 240 241 242 243 244 245 246

  /**
   * Allocates memory for store
   * @throws bad_alloc: when there is not enough memory or when the
   *                    the total number of kmers cannot be stored in an 
   *                    unsigned int
   */
  void init();
Mikaël Salson's avatar
Mikaël Salson committed
247 248 249 250 251 252 253 254
public:
  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  ArrayKmerStore(string seed, bool=false);
  ArrayKmerStore(int k, bool=false);
  ~ArrayKmerStore();
255

256
  vector<T> getResults(const seqtype &seq, bool no_revcomp=false, const string seed="");
257 258
  T& get(seqtype &word);
  T& operator[](seqtype & word);
259
  T& operator[](int word);
Mikaël Salson's avatar
Mikaël Salson committed
260 261 262 263 264 265 266 267 268
};


// IKmerStore

template<class T>
IKmerStore<T>::~IKmerStore(){}

template<class T> 
269
void IKmerStore<T>::insert(list<BioReader>& input,
270
                           const string &label,
271 272
                           int keep_only,
                           string seed){
273
  for(list<BioReader>::iterator it = input.begin() ; it != input.end() ; it++){
274
    insert(*it, label, keep_only, seed);
Mikaël Salson's avatar
Mikaël Salson committed
275 276 277 278
  }
}

template<class T> 
279
void IKmerStore<T>::insert(BioReader& input,
280
                           const string &label,
281 282
                           int keep_only,
                           string seed){
Mikaël Salson's avatar
Mikaël Salson committed
283
  for (int r = 0; r < input.size(); r++) {
284
    insert(input.sequence(r), label, true, keep_only, seed);
Mikaël Salson's avatar
Mikaël Salson committed
285
  }
286

287
  labels.push_back(make_pair(T(label, 1, seed.size()), input)) ;
288 289

  if (revcomp_indexed  && ! T::hasRevcompSymetry()) {
290
    labels.push_back(make_pair(T(label, -1, seed.size()), input)) ;
291
  }
Mikaël Salson's avatar
Mikaël Salson committed
292 293 294
}

template<class T> 
295
void IKmerStore<T>::insert(const seqtype &sequence,
296
                           const string &label,
297
                           bool ignore_extended_nucleotides,
298 299
                           int keep_only,
                           string seed){
300 301 302 303 304 305 306
  size_t start_indexing = 0;
  size_t end_indexing = sequence.length();
  if (keep_only > 0 && sequence.length() > (size_t)keep_only) {
    start_indexing = sequence.length() - keep_only;
  } else if (keep_only < 0 && sequence.length() > (size_t) -keep_only) {
    end_indexing = -keep_only;
  }
307

308
  if (seed.empty() || ! this->hasDifferentKmerTypes())
309 310 311
    seed = this->seed;

  size_t seed_span = seed.length();
312 313 314 315 316
  size_t size_indexing = end_indexing - start_indexing;
  if (size_indexing > max_size_indexing) {
    max_size_indexing = size_indexing;
  }

317 318
  for(size_t i = start_indexing ; i + seed_span < end_indexing + 1 ; i++) {
    seqtype substr = sequence.substr(i, seed_span);
319
    seqtype kmer = spaced(substr, seed);
320

321
    if (ignore_extended_nucleotides && has_extended_nucleotides(kmer))
322 323
      continue;

Mikaël Salson's avatar
Mikaël Salson committed
324 325
    int strand = 1;
    if (revcomp_indexed && T::hasRevcompSymetry()) {
326
      seqtype rc_kmer = revcomp(kmer);
Mikaël Salson's avatar
Mikaël Salson committed
327 328 329 330 331
      if (rc_kmer.compare(kmer) < 0) {
        strand = -1;
        kmer = rc_kmer;
      }
    }
332 333 334 335
    T &this_kmer = this->get(kmer);
    if (this_kmer.isNull()) {
      nb_kmers_inserted++;
    }
336
    this_kmer += T(label, strand, seed.size());
Mikaël Salson's avatar
Mikaël Salson committed
337
    if (revcomp_indexed && ! T::hasRevcompSymetry()) {
338
      seqtype rc_kmer = spaced(revcomp(substr), seed);
339 340 341
      T &this_rc_kmer = this->get(rc_kmer);
      if (this_rc_kmer.isNull())
        nb_kmers_inserted++;
342
      this_rc_kmer += T(label, -1, seed.size());
Mikaël Salson's avatar
Mikaël Salson committed
343 344 345
    }
  }
}
346

347
template<class T>
348 349 350
void IKmerStore<T>::finish_building() {
  finished_building = true;
}
351

352 353
template<class T>
float IKmerStore<T>::getIndexLoad(const T kmer) const {
354
  float index_load = nb_kmers_inserted / pow(4.0, k);
355
  return (kmer.isUnknown()) ? 1 - index_load : index_load;
356
}
Mikaël Salson's avatar
Mikaël Salson committed
357

358 359 360 361 362 363 364
template<class T>
int IKmerStore<T>::atMostMaxSizeIndexing(int n) const {
  if (!max_size_indexing || n < max_size_indexing)
    return n ;

  return max_size_indexing ;
}
365 366

template<class T>
367
double IKmerStore<T>::getProbabilityAtLeastOrAbove(const T kmer, int at_least, int length) const {
368

369 370
  if (at_least == 0) return 1.0; // even if 'length' is very small
  
371 372
  // n: number of kmers in the sequence
  int n = length - getS() + 1;
373
  float index_load = getIndexLoad(kmer) ;
374 375 376 377 378 379 380 381 382 383

  double proba = 0;
  double probability_having_system = pow(index_load, at_least);
  double probability_not_having_system = pow(1 - index_load, n - at_least);
  for (int i=at_least; i<=n; i++) {
    proba += nChoosek(n, i) * probability_having_system * probability_not_having_system;
    probability_having_system *= index_load;
    probability_not_having_system /= (1 - index_load);
  }

384 385 386 387
#ifdef DEBUG_KMS_EVALUE
  cerr << "e-value:\tindex_load=" << index_load << ",\tat_least=" << at_least << ",\tlength=" << length <<",\tp-value=" << proba << endl;
#endif

388 389 390 391 392
  return proba;
}



Mikaël Salson's avatar
Mikaël Salson committed
393 394 395 396 397 398 399 400 401 402
template<class T>
int IKmerStore<T>::getK() const {
  return k;
}

template<class T>
int IKmerStore<T>::getS() const {
  return s;
}

403 404 405 406 407
template<class T>
string IKmerStore<T>::getSeed() const {
  return seed;
}

408
template<class T>
409 410
BioReader IKmerStore<T>::getLabel(T kmer) const {
  for (typename list< pair<T, BioReader> >::const_iterator it = labels.begin(); it != labels.end(); ++it)
411 412
    if (it->first == kmer)
      return it->second ;
413 414 415 416 417 418
  // Nothing interesting found
  // Try by ignoring length if the index is not able to deal with different lengths
  if (! hasDifferentKmerTypes() && kmer.getLength() != (unsigned char)~0) {
    kmer.setLength(~0);
    return getLabel(kmer);
  }
419
  return BIOREADER_AMBIGUOUS ;
420
}
421

422 423 424 425 426
template<class T>
bool IKmerStore<T>::hasDifferentKmerTypes() const {
  return false;
}

427
// .getResults()
Mikaël Salson's avatar
Mikaël Salson committed
428
template<class T>
429
vector<T> MapKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp, string seed) {
430

431
  string local_seed = seed;
432
  if (! seed.size())
433 434
    local_seed = IKmerStore<T>::seed;
  int s = local_seed.size();
435

Mikaël Salson's avatar
Mikaël Salson committed
436 437 438 439 440
  if ((int)seq.length() < s - 1) {
    return vector<T>(0);
  }
  vector<T> result(seq.length() - s + 1);
  for (size_t i=0; i + s < seq.length() + 1; i++) {
441
    seqtype kmer = spaced(seq.substr(i, s), local_seed);
442
    //    seqtype kmer = seq.substr(i, s);
Mikaël Salson's avatar
Mikaël Salson committed
443
    // cout << kmer << endl << kmer0 << endl << endl ;
444
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
Mikaël Salson's avatar
Mikaël Salson committed
445 446 447 448 449 450 451 452
      result[i] = get(kmer);
    } else {
      result[i] = (*this)[kmer];
    }
  }
  return result;
}

453
template<class T>
454
vector<T> ArrayKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp, string seed) {
455 456

  string local_seed = seed;
457
  if (! seed.size())
458 459
    local_seed = IKmerStore<T>::seed;
  int s = local_seed.size();
460 461 462 463 464 465 466 467 468 469 470 471

  int N = (int)seq.length();

  if (N < s - 1) {
    return vector<T>(0);
  }
  vector<T> result(N - s + 1);

  /* Read once the sequence, convert it to int* */
  int* intseq = new int[N];
  for (int i=0; i<N; i++)
    {
472
      intseq[i] = nuc_to_int(seq[i]);
473 474 475 476
    }

  /* Compute results */
  for (size_t i=0; (int) i+s < N+1; i++) {
477
    int kmer = spaced_int(intseq + i, local_seed);
478 479 480 481
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
      result[i] = store[kmer]; // getfromint(kmer); // store[kmer];
      // cout << i << "/" << N << "  " << kmer << result[i] << endl ;
    } else {
482
      result[i] = (*this)[kmer]; // Deals with revcomp
483 484 485 486 487 488 489
    }
  }

  delete[] intseq ;
  return result;
}

Mikaël Salson's avatar
Mikaël Salson committed
490 491 492 493 494 495 496 497 498

template<class T>
bool IKmerStore<T>::isRevcomp() const {
  return revcomp_indexed;
}

// MapKmerStore

template <class T>
499
MapKmerStore<T>::MapKmerStore(string seed, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
500 501 502 503 504
  this->seed = seed;   
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
505
  init();
Mikaël Salson's avatar
Mikaël Salson committed
506 507 508
}

template <class T>
509
MapKmerStore<T>::MapKmerStore(int k, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
510 511 512 513
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
514 515 516 517 518 519
  init();
}

template <class T>
void MapKmerStore<T>::init() {
  this->nb_kmers_inserted = 0;
520
  this->max_size_indexing = 0;
Mikaël Salson's avatar
Mikaël Salson committed
521 522 523
}

template <class T> 
524
T& MapKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
525 526 527 528
  return store[word];
}

template <class T> 
529
T& MapKmerStore<T>::operator[](seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
530
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
531
    seqtype rc_kmer = revcomp(word);
Mikaël Salson's avatar
Mikaël Salson committed
532 533 534 535 536 537 538 539 540
    if (rc_kmer.compare(word) < 0)
      word = rc_kmer;
  }
  return store[word];
}

// ArrayKmerStore

template <class T> 
541
ArrayKmerStore<T>::ArrayKmerStore(int k, bool revcomp):IKmerStore<T>() {
Mikaël Salson's avatar
Mikaël Salson committed
542 543 544 545
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
546
  init();
Mikaël Salson's avatar
Mikaël Salson committed
547 548 549 550
}


template <class T> 
551
ArrayKmerStore<T>::ArrayKmerStore(string seed, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
552 553 554 555 556
  this->seed = seed; 
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
557 558 559 560 561
  init();
}

template <class T>
void ArrayKmerStore<T>::init() {
562
  this->nb_kmers_inserted = 0;
563
  this->max_size_indexing = 0;
564
  if ((size_t)(this->k << 1) >= sizeof(int) * 8)
Mikaël Salson's avatar
Mikaël Salson committed
565
    throw std::bad_alloc();
566
  store = new T[(unsigned int)1 << (this->k << 1)];
Mikaël Salson's avatar
Mikaël Salson committed
567 568 569 570 571 572 573 574 575 576 577 578 579 580
  if (! store)
    throw std::bad_alloc();
}

template <class T> 
ArrayKmerStore<T>::~ArrayKmerStore(){
  delete [] store;
}

/**	Returns index of word in an array of size 4^k
	Considering B(A) = 0, B(C) = 1, B(T) = 2, B(G) = 3
	index_word = \sum_{i=0}^{k-1}B(word[i])*4^(k-i-1)
**/
template <class T> 
581
int ArrayKmerStore<T>::index(const seqtype& word) const{
Mikaël Salson's avatar
Mikaël Salson committed
582
  return dna_to_int(word, this->k);
Mikaël Salson's avatar
Mikaël Salson committed
583 584 585
}

template <class T> 
586
T& ArrayKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
587 588 589 590
  return store[index(word)];
}

template <class T> 
591
T& ArrayKmerStore<T>::operator[](seqtype& word){
592 593 594 595 596
  return (*this)[index(word)];
}

template <class T> 
T& ArrayKmerStore<T>::operator[](int word){
Mikaël Salson's avatar
Mikaël Salson committed
597
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
598 599
    int rc_kmer = revcomp_int(word, IKmerStore<T>::k);
    if (rc_kmer < word)
Mikaël Salson's avatar
Mikaël Salson committed
600 601
      word = rc_kmer;
  }
602
  return store[word];
Mikaël Salson's avatar
Mikaël Salson committed
603 604 605
}

#endif