kmerstore.h 15.6 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>
Mikaël Salson's avatar
Mikaël Salson committed
10 11 12 13 14 15 16 17 18 19
#include "fasta.h"
#include "tools.h"

using namespace std;

class Kmer {
public:
  unsigned int count;

  Kmer();
20 21 22 23

  /**
   * This constructor is used via a IKmerStore<Kmer> index (hence the argument list)       
   */
24
  Kmer(const string &label, int strand=1, size_t length=0);
25

Mikaël Salson's avatar
Mikaël Salson committed
26
  Kmer &operator+=(const Kmer &);
27

28 29 30 31 32
  /**
   * @return the length of the kmer (only there for compatibility with KmerAffect)
   */
  size_t getLength() const;

33 34 35 36
  /**
   * 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
37
  static bool hasRevcompSymetry();
38 39 40 41

  /**
   * @return true if the element is the same as when initialised with default constructor.
   */
42 43 44 45 46 47 48
  bool isNull() const;

  /**
   * @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
49 50 51 52 53 54 55 56 57 58 59 60 61
} ;
ostream &operator<<(ostream &os, const Kmer &kmer);


/* 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 ;
62
  size_t nb_kmers_inserted;
63
  size_t max_size_indexing;
64
  bool finished_building;
Mikaël Salson's avatar
Mikaël Salson committed
65 66 67 68 69

public:

  virtual ~IKmerStore();

70 71
  static int last_id;
  int id; // id of this index
72
  int refs; // number of germlines using this index
73

74
  list< pair <T, Fasta> > labels;
75

76 77
  IKmerStore();

Mikaël Salson's avatar
Mikaël Salson committed
78 79 80
  /**
   * @param input: A single FASTA file
   * @param label: label that must be associated to the given files
81
   * @param seed: the seed to use for indexing. By default it will be the seed of the index
82
   * @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
83
   */
84
  void insert(Fasta& input, const string& label="", int keep_only = 0, string seed = "");
Mikaël Salson's avatar
Mikaël Salson committed
85 86 87 88

  /**
   * @param input: A list of FASTA files
   * @param label: label that must be associated to the given files
89
   * @param seed: the seed to use for indexing. By default it will be the seed of the index
90
   * @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
91
   */
92
  void insert(list<Fasta>& input, const string& label="", int keep_only = 0, string seed = "");
Mikaël Salson's avatar
Mikaël Salson committed
93 94 95 96
  
  /**
   * @param input: A sequence to be cut in k-mers
   * @param label: label that must be associated to the given files
97 98 99 100
   * @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.
101
   * @param seed: the seed to use for indexing. By default it will be the seed of the index
Mikaël Salson's avatar
Mikaël Salson committed
102 103
   * @post All the k-mers in the sequence have been indexed.
   */
104 105 106 107 108 109 110 111 112 113
  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.
   */
114
  virtual void finish_building();
Mikaël Salson's avatar
Mikaël Salson committed
115 116 117 118 119

  /**
   * @param word: a k-mer
   * @return the value for the kmer as stored in the structure (don't do revcomp)
   */ 
120
  virtual T& get(seqtype &word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
121

122 123 124
  /**
   * @return the percentage of kmers that are set in the index
   */
125
  virtual float getIndexLoad(T kmer) const;
126

127 128 129 130 131 132
  /**
   * @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;

133 134 135
  /**
   * @return probability that the number of kmers is 'at_least' or more in a sequence of length 'length'
   */
136
  double getProbabilityAtLeastOrAbove(T kmer, int at_least, int length) const;
137

Mikaël Salson's avatar
Mikaël Salson committed
138 139 140 141 142 143 144 145 146 147
  /**
   * @return the value of k
   */
  int getK() const;

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

148 149 150 151 152
  /**
   * @return the seed used
   */
  string getSeed() const;

153 154 155 156
  /**
   * @param kmer: a kmer
   * @return one label associated with the kmer
   */
157
  Fasta getLabel(T kmer) const;
158

Mikaël Salson's avatar
Mikaël Salson committed
159 160 161 162
  /**
   * @param seq: a sequence
   * @param no_revcomp: force not to revcomp the sequence, even if
   *                    the index was built with revcomp.
163 164
   * @return a vector of length seq.length() - smallestAnalysableLength() + 1
   * containing for each position the corresponding value in the index.
Mikaël Salson's avatar
Mikaël Salson committed
165
   */
166
  virtual vector<T> getResults(const seqtype &seq, bool no_revcomp=false, string seed="") = 0;
Mikaël Salson's avatar
Mikaël Salson committed
167

168 169 170 171 172 173 174
  /**
   * @return the smallest analysable sequence length by the index. For an
   * index just storing k-mers it will obviously be getK(). But for more
   * sophisticated data structures it could be 1 for instance.
   */
  size_t smallestAnalysableLength() const;

Mikaël Salson's avatar
Mikaël Salson committed
175 176 177 178 179 180 181 182 183 184 185
  /**
   * @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
   */
186
  virtual T& operator[](seqtype& word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
187 188
};

189 190 191
template<class T>
IKmerStore<T>::IKmerStore() {
  id = ++last_id;
Mikaël Salson's avatar
Mikaël Salson committed
192
  refs = 0;
193
  finished_building = false;
194 195
}

196
template<class T> int IKmerStore<T>::last_id = 0;
Mikaël Salson's avatar
Mikaël Salson committed
197 198 199 200 201

template <class T> 
class MapKmerStore : public IKmerStore<T>
{
public:	
202
  map<seqtype, T> store;
Mikaël Salson's avatar
Mikaël Salson committed
203 204 205 206 207 208 209

  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  MapKmerStore(string seed, bool=false);
  MapKmerStore(int k, bool=false);
210

211
  vector<T> getResults(const seqtype &seq, bool no_revcomp=false, string seed="");
212 213
  T& get(seqtype &word);
  T& operator[](seqtype & word);
214 215 216

 private:
  void init();
Mikaël Salson's avatar
Mikaël Salson committed
217 218 219 220 221 222 223
};

template <class T> 
class ArrayKmerStore : public IKmerStore<T>
{
  T* store;
	
224
  int index(const seqtype& word) const;
225 226 227 228 229 230 231 232

  /**
   * 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
233 234 235 236 237 238 239 240
public:
  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  ArrayKmerStore(string seed, bool=false);
  ArrayKmerStore(int k, bool=false);
  ~ArrayKmerStore();
241

242
  vector<T> getResults(const seqtype &seq, bool no_revcomp=false, const string seed="");
243 244
  T& get(seqtype &word);
  T& operator[](seqtype & word);
245
  T& operator[](int word);
Mikaël Salson's avatar
Mikaël Salson committed
246 247 248 249 250 251 252 253 254 255
};


// IKmerStore

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

template<class T> 
void IKmerStore<T>::insert(list<Fasta>& input,
256
                           const string &label,
257 258
                           int keep_only,
                           string seed){
Mikaël Salson's avatar
Mikaël Salson committed
259
  for(list<Fasta>::iterator it = input.begin() ; it != input.end() ; it++){
260
    insert(*it, label, keep_only, seed);
Mikaël Salson's avatar
Mikaël Salson committed
261 262 263 264 265
  }
}

template<class T> 
void IKmerStore<T>::insert(Fasta& input,
266
                           const string &label,
267 268
                           int keep_only,
                           string seed){
Mikaël Salson's avatar
Mikaël Salson committed
269
  for (int r = 0; r < input.size(); r++) {
270
    insert(input.sequence(r), label, true, keep_only, seed);
Mikaël Salson's avatar
Mikaël Salson committed
271
  }
272

273
  labels.push_back(make_pair(T(label, 1, seed_weight(seed)), input)) ;
274 275

  if (revcomp_indexed  && ! T::hasRevcompSymetry()) {
276
    labels.push_back(make_pair(T(label, -1, seed_weight(seed)), input)) ;
277
  }
Mikaël Salson's avatar
Mikaël Salson committed
278 279 280
}

template<class T> 
281
void IKmerStore<T>::insert(const seqtype &sequence,
282
                           const string &label,
283
                           bool ignore_extended_nucleotides,
284 285
                           int keep_only,
                           string seed){
286 287 288 289 290 291 292
  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;
  }
293

294 295 296 297
  if (seed.empty())
    seed = this->seed;

  size_t seed_span = seed.length();
298
  size_t seed_w = seed_weight(seed);
299 300 301 302 303
  size_t size_indexing = end_indexing - start_indexing;
  if (size_indexing > max_size_indexing) {
    max_size_indexing = size_indexing;
  }

304 305
  for(size_t i = start_indexing ; i + seed_span < end_indexing + 1 ; i++) {
    seqtype substr = sequence.substr(i, seed_span);
306
    seqtype kmer = spaced(substr, seed);
307

308
    if (ignore_extended_nucleotides && has_extended_nucleotides(kmer))
309 310
      continue;

Mikaël Salson's avatar
Mikaël Salson committed
311 312
    int strand = 1;
    if (revcomp_indexed && T::hasRevcompSymetry()) {
313
      seqtype rc_kmer = revcomp(kmer);
Mikaël Salson's avatar
Mikaël Salson committed
314 315 316 317 318
      if (rc_kmer.compare(kmer) < 0) {
        strand = -1;
        kmer = rc_kmer;
      }
    }
319 320 321 322
    T &this_kmer = this->get(kmer);
    if (this_kmer.isNull()) {
      nb_kmers_inserted++;
    }
323
    this_kmer += T(label, strand, seed_w);
Mikaël Salson's avatar
Mikaël Salson committed
324
    if (revcomp_indexed && ! T::hasRevcompSymetry()) {
325
      seqtype rc_kmer = spaced(revcomp(substr), seed);
326 327 328
      T &this_rc_kmer = this->get(rc_kmer);
      if (this_rc_kmer.isNull())
        nb_kmers_inserted++;
329
      this_rc_kmer += T(label, -1, seed_w);
Mikaël Salson's avatar
Mikaël Salson committed
330 331 332
    }
  }
}
333

334
template<class T>
335 336 337
void IKmerStore<T>::finish_building() {
  finished_building = true;
}
338

339 340
template<class T>
float IKmerStore<T>::getIndexLoad(const T kmer) const {
341 342
  float index_load = nb_kmers_inserted / pow(4.0, kmer.getLength());
  return (kmer.isUnknown()) ? 1 - index_load : index_load;
343
}
Mikaël Salson's avatar
Mikaël Salson committed
344

345 346 347 348 349 350 351
template<class T>
int IKmerStore<T>::atMostMaxSizeIndexing(int n) const {
  if (!max_size_indexing || n < max_size_indexing)
    return n ;

  return max_size_indexing ;
}
352 353

template<class T>
354
double IKmerStore<T>::getProbabilityAtLeastOrAbove(const T kmer, int at_least, int length) const {
355 356 357

  // n: number of kmers in the sequence
  int n = length - getS() + 1;
358
  float index_load = getIndexLoad(kmer) ;
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373

  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);
  }

  return proba;
}



Mikaël Salson's avatar
Mikaël Salson committed
374 375 376 377 378 379 380 381 382 383
template<class T>
int IKmerStore<T>::getK() const {
  return k;
}

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

384 385 386 387 388
template<class T>
string IKmerStore<T>::getSeed() const {
  return seed;
}

389
template<class T>
390 391
Fasta IKmerStore<T>::getLabel(T kmer) const {
  for (typename list< pair<T, Fasta> >::const_iterator it = labels.begin(); it != labels.end(); ++it)
392 393
    if (it->first == kmer)
      return it->second ;
394
  return FASTA_AMBIGUOUS ;
395
}
396 397

// .getResults()
Mikaël Salson's avatar
Mikaël Salson committed
398
template<class T>
399
vector<T> MapKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp, string seed) {
400

Mikaël Salson's avatar
Mikaël Salson committed
401
  string local_seed = seed;
402
  if (! seed.size())
Mikaël Salson's avatar
Mikaël Salson committed
403 404
    local_seed = IKmerStore<T>::seed;
  int s = local_seed.size();
405

Mikaël Salson's avatar
Mikaël Salson committed
406 407 408 409 410
  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++) {
Mikaël Salson's avatar
Mikaël Salson committed
411
    seqtype kmer = spaced(seq.substr(i, s), local_seed);
412
    //    seqtype kmer = seq.substr(i, s);
Mikaël Salson's avatar
Mikaël Salson committed
413
    // cout << kmer << endl << kmer0 << endl << endl ;
414
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
Mikaël Salson's avatar
Mikaël Salson committed
415 416 417 418 419 420 421 422
      result[i] = get(kmer);
    } else {
      result[i] = (*this)[kmer];
    }
  }
  return result;
}

423
template<class T>
424
vector<T> ArrayKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp, string seed) {
Mikaël Salson's avatar
Mikaël Salson committed
425 426

  string local_seed = seed;
427
  if (! seed.size())
Mikaël Salson's avatar
Mikaël Salson committed
428 429
    local_seed = IKmerStore<T>::seed;
  int s = local_seed.size();
430 431 432 433 434 435 436 437 438 439 440 441

  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++)
    {
442
      intseq[i] = nuc_to_int(seq[i]);
443 444 445 446
    }

  /* Compute results */
  for (size_t i=0; (int) i+s < N+1; i++) {
Mikaël Salson's avatar
Mikaël Salson committed
447
    int kmer = spaced_int(intseq + i, local_seed);
448 449 450 451
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
      result[i] = store[kmer]; // getfromint(kmer); // store[kmer];
      // cout << i << "/" << N << "  " << kmer << result[i] << endl ;
    } else {
452
      result[i] = (*this)[kmer]; // Deals with revcomp
453 454 455 456 457 458 459
    }
  }

  delete[] intseq ;
  return result;
}

Mikaël Salson's avatar
Mikaël Salson committed
460 461 462 463 464 465

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

466 467 468 469 470
template<class T>
size_t IKmerStore<T>::smallestAnalysableLength() const {
  return getK();
}

Mikaël Salson's avatar
Mikaël Salson committed
471 472 473
// MapKmerStore

template <class T>
474
MapKmerStore<T>::MapKmerStore(string seed, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
475 476 477 478 479
  this->seed = seed;   
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
480
  init();
Mikaël Salson's avatar
Mikaël Salson committed
481 482 483
}

template <class T>
484
MapKmerStore<T>::MapKmerStore(int k, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
485 486 487 488
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
489 490 491 492 493 494
  init();
}

template <class T>
void MapKmerStore<T>::init() {
  this->nb_kmers_inserted = 0;
495
  this->max_size_indexing = 0;
Mikaël Salson's avatar
Mikaël Salson committed
496 497 498
}

template <class T> 
499
T& MapKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
500 501 502 503
  return store[word];
}

template <class T> 
504
T& MapKmerStore<T>::operator[](seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
505
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
506
    seqtype rc_kmer = revcomp(word);
Mikaël Salson's avatar
Mikaël Salson committed
507 508 509 510 511 512 513 514 515
    if (rc_kmer.compare(word) < 0)
      word = rc_kmer;
  }
  return store[word];
}

// ArrayKmerStore

template <class T> 
516
ArrayKmerStore<T>::ArrayKmerStore(int k, bool revcomp):IKmerStore<T>() {
Mikaël Salson's avatar
Mikaël Salson committed
517 518 519 520
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
521
  init();
Mikaël Salson's avatar
Mikaël Salson committed
522 523 524 525
}


template <class T> 
526
ArrayKmerStore<T>::ArrayKmerStore(string seed, bool revcomp):IKmerStore<T>(){
Mikaël Salson's avatar
Mikaël Salson committed
527 528 529 530 531
  this->seed = seed; 
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
532 533 534 535 536
  init();
}

template <class T>
void ArrayKmerStore<T>::init() {
537
  this->nb_kmers_inserted = 0;
538
  this->max_size_indexing = 0;
539
  if ((size_t)(this->k << 1) >= sizeof(int) * 8)
Mikaël Salson's avatar
Mikaël Salson committed
540
    throw std::bad_alloc();
541
  store = new T[(unsigned int)1 << (this->k << 1)];
Mikaël Salson's avatar
Mikaël Salson committed
542 543 544 545 546 547 548 549 550 551 552 553 554 555
  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> 
556
int ArrayKmerStore<T>::index(const seqtype& word) const{
Mikaël Salson's avatar
Mikaël Salson committed
557
  return dna_to_int(word, this->k);
Mikaël Salson's avatar
Mikaël Salson committed
558 559 560
}

template <class T> 
561
T& ArrayKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
562 563 564 565
  return store[index(word)];
}

template <class T> 
566
T& ArrayKmerStore<T>::operator[](seqtype& word){
567 568 569 570 571
  return (*this)[index(word)];
}

template <class T> 
T& ArrayKmerStore<T>::operator[](int word){
Mikaël Salson's avatar
Mikaël Salson committed
572
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
573 574
    int rc_kmer = revcomp_int(word, IKmerStore<T>::k);
    if (rc_kmer < word)
Mikaël Salson's avatar
Mikaël Salson committed
575 576
      word = rc_kmer;
  }
577
  return store[word];
Mikaël Salson's avatar
Mikaël Salson committed
578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598
}


/**
 * KmerStoreFactory is a factory that allows to create an index that best fits
 * your needs!
 */
class KmerStoreFactory {
 public:
  template<class T>
    static IKmerStore<T> *createIndex(string seed, bool revcomp=false);
  template<class T>
    static IKmerStore<T> *createIndex(int k, bool revcomp=false);
};

template<class T>
IKmerStore<T> *KmerStoreFactory::createIndex(string seed, bool revcomp) {
  IKmerStore<T> *index;
  try{
    index = new ArrayKmerStore<T>(seed, revcomp);
  }catch(exception e){
599
    cout << "  (using a MapKmer to fit into memory)" << endl;
Mikaël Salson's avatar
Mikaël Salson committed
600 601
    index = new MapKmerStore<T>(seed, revcomp);
  }
602

Mikaël Salson's avatar
Mikaël Salson committed
603 604 605 606 607 608 609 610 611
  return index;
}

template<class T>
IKmerStore<T> *KmerStoreFactory::createIndex(int k, bool revcomp) {
  return createIndex<T>(seed_contiguous(k), revcomp);
}

#endif