kmerstore.h 11.2 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);
25

Mikaël Salson's avatar
Mikaël Salson committed
26 27
  Kmer &operator+=(const Kmer &);
  static bool hasRevcompSymetry();
28 29 30 31 32

  /**
   * @return true if the element is the same as when initialised with default constructor.
   */
  bool isNull();
Mikaël Salson's avatar
Mikaël Salson committed
33 34 35 36 37 38 39 40 41 42 43 44 45
} ;
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 ;
46
  size_t nb_kmers_inserted;
Mikaël Salson's avatar
Mikaël Salson committed
47 48 49 50 51

public:

  virtual ~IKmerStore();

52 53
  list< pair <T, string> > labels;

Mikaël Salson's avatar
Mikaël Salson committed
54 55 56
  /**
   * @param input: A single FASTA file
   * @param label: label that must be associated to the given files
57
   * @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
58 59 60 61 62 63
   */
  void insert(Fasta& input, const string& label="");

  /**
   * @param input: A list of FASTA files
   * @param label: label that must be associated to the given files
64
   * @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
65 66 67 68 69 70 71 72
   */
  void insert(list<Fasta>& input, const string& label="");
  
  /**
   * @param input: A sequence to be cut in k-mers
   * @param label: label that must be associated to the given files
   * @post All the k-mers in the sequence have been indexed.
   */
73
  void insert(const seqtype &sequence,
74 75
              const string &label,
              bool ignore_extended_nucleotides=true);
Mikaël Salson's avatar
Mikaël Salson committed
76 77 78 79 80

  /**
   * @param word: a k-mer
   * @return the value for the kmer as stored in the structure (don't do revcomp)
   */ 
81
  virtual T& get(seqtype &word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
82

83 84 85 86 87
  /**
   * @return the percentage of kmers that are set in the index
   */
  float getIndexLoad() const;

88 89 90 91 92
  /**
   * @return probability that the number of kmers is 'at_least' or more in a sequence of length 'length'
   */
  double getProbabilityAtLeastOrAbove(int at_least, int length) const;

Mikaël Salson's avatar
Mikaël Salson committed
93 94 95 96 97 98 99 100 101 102
  /**
   * @return the value of k
   */
  int getK() const;

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

103 104 105 106 107
  /**
   * @return the seed used
   */
  string getSeed() const;

Mikaël Salson's avatar
Mikaël Salson committed
108 109 110 111 112 113 114
  /**
   * @param seq: a sequence
   * @param no_revcomp: force not to revcomp the sequence, even if
   *                    the index was built with revcomp.
   * @return a vector of length seq.length() - getK() + 1 containing
   * for each k-mer the corresponding value in the index.
   */
115
  virtual vector<T> getResults(const seqtype &seq, bool no_revcomp=false) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
116 117 118 119 120 121 122 123 124 125 126 127

  /**
   * @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
   */
128
  virtual T& operator[](seqtype& word) = 0;
Mikaël Salson's avatar
Mikaël Salson committed
129 130 131 132 133 134 135
};


template <class T> 
class MapKmerStore : public IKmerStore<T>
{
public:	
136
  map<seqtype, T> store;
Mikaël Salson's avatar
Mikaël Salson committed
137 138 139 140 141 142 143

  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  MapKmerStore(string seed, bool=false);
  MapKmerStore(int k, bool=false);
144 145

  vector<T> getResults(const seqtype &seq, bool no_revcomp=false);
146 147
  T& get(seqtype &word);
  T& operator[](seqtype & word);
148 149 150

 private:
  void init();
Mikaël Salson's avatar
Mikaël Salson committed
151 152 153 154 155 156 157
};

template <class T> 
class ArrayKmerStore : public IKmerStore<T>
{
  T* store;
	
158
  int index(const seqtype& word) const;
159 160 161 162 163 164 165 166

  /**
   * 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
167 168 169 170 171 172 173 174
public:
  /**
   * @param bool: if true, will also index revcomp
   * @param seed: the seed
   */
  ArrayKmerStore(string seed, bool=false);
  ArrayKmerStore(int k, bool=false);
  ~ArrayKmerStore();
175 176

  vector<T> getResults(const seqtype &seq, bool no_revcomp=false);	
177 178
  T& get(seqtype &word);
  T& operator[](seqtype & word);
179
  T& operator[](int word);
Mikaël Salson's avatar
Mikaël Salson committed
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
};


// IKmerStore

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

template<class T> 
void IKmerStore<T>::insert(list<Fasta>& input,
                           const string &label){
  for(list<Fasta>::iterator it = input.begin() ; it != input.end() ; it++){
    insert(*it, label);
  }
}

template<class T> 
void IKmerStore<T>::insert(Fasta& input,
                           const string &label){
  for (int r = 0; r < input.size(); r++) {
    insert(input.sequence(r), label);
  }
202

203
  labels.push_back(make_pair(T(label, 1), label)) ;
Mikaël Salson's avatar
Mikaël Salson committed
204 205 206
}

template<class T> 
207
void IKmerStore<T>::insert(const seqtype &sequence,
208 209
                           const string &label,
                           bool ignore_extended_nucleotides){
Mikaël Salson's avatar
Mikaël Salson committed
210
  for(size_t i = 0 ; i + s < sequence.length() + 1 ; i++) {
211
    seqtype kmer = spaced(sequence.substr(i, s), seed);
212

213
    if (ignore_extended_nucleotides && has_extended_nucleotides(kmer))
214 215
      continue;

Mikaël Salson's avatar
Mikaël Salson committed
216 217
    int strand = 1;
    if (revcomp_indexed && T::hasRevcompSymetry()) {
218
      seqtype rc_kmer = revcomp(kmer);
Mikaël Salson's avatar
Mikaël Salson committed
219 220 221 222 223
      if (rc_kmer.compare(kmer) < 0) {
        strand = -1;
        kmer = rc_kmer;
      }
    }
224 225 226 227 228
    T &this_kmer = this->get(kmer);
    if (this_kmer.isNull()) {
      nb_kmers_inserted++;
    }
    this_kmer += T(label, strand);
Mikaël Salson's avatar
Mikaël Salson committed
229
    if (revcomp_indexed && ! T::hasRevcompSymetry()) {
230
      seqtype rc_kmer = revcomp(kmer);
231 232 233 234
      T &this_rc_kmer = this->get(rc_kmer);
      if (this_rc_kmer.isNull())
        nb_kmers_inserted++;
      this_rc_kmer += T(label, -1);
Mikaël Salson's avatar
Mikaël Salson committed
235 236 237
    }
  }
}
238 239 240 241 242

template<class T>
float IKmerStore<T>::getIndexLoad() const {
  return nb_kmers_inserted*1. / (1 << (2 * k));
}
Mikaël Salson's avatar
Mikaël Salson committed
243

244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265

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

  // n: number of kmers in the sequence
  int n = length - getS() + 1;
  float index_load = getIndexLoad() ;

  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
266 267 268 269 270 271 272 273 274 275
template<class T>
int IKmerStore<T>::getK() const {
  return k;
}

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

276 277 278 279 280
template<class T>
string IKmerStore<T>::getSeed() const {
  return seed;
}

281 282

// .getResults()
Mikaël Salson's avatar
Mikaël Salson committed
283
template<class T>
284 285 286 287
vector<T> MapKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp) {

  int s = IKmerStore<T>::getS();

Mikaël Salson's avatar
Mikaël Salson committed
288 289 290 291 292
  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++) {
293
    seqtype kmer = spaced(seq.substr(i, s), IKmerStore<T>::seed);
294
    //    seqtype kmer = seq.substr(i, s);
Mikaël Salson's avatar
Mikaël Salson committed
295
    // cout << kmer << endl << kmer0 << endl << endl ;
296
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
Mikaël Salson's avatar
Mikaël Salson committed
297 298 299 300 301 302 303 304
      result[i] = get(kmer);
    } else {
      result[i] = (*this)[kmer];
    }
  }
  return result;
}

305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
template<class T>
vector<T> ArrayKmerStore<T>::getResults(const seqtype &seq, bool no_revcomp) {
  
  int s = IKmerStore<T>::getS();

  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++)
    {
321
      intseq[i] = nuc_to_int(seq[i]);
322 323 324 325 326 327 328 329 330
    }

  /* Compute results */
  for (size_t i=0; (int) i+s < N+1; i++) {
    int kmer = spaced_int(intseq + i, IKmerStore<T>::seed);
    if (IKmerStore<T>::revcomp_indexed && no_revcomp) {
      result[i] = store[kmer]; // getfromint(kmer); // store[kmer];
      // cout << i << "/" << N << "  " << kmer << result[i] << endl ;
    } else {
331
      result[i] = (*this)[kmer]; // Deals with revcomp
332 333 334 335 336 337 338
    }
  }

  delete[] intseq ;
  return result;
}

Mikaël Salson's avatar
Mikaël Salson committed
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353

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

// MapKmerStore

template <class T>
MapKmerStore<T>::MapKmerStore(string seed, bool revcomp){
  this->seed = seed;   
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
354
  init();
Mikaël Salson's avatar
Mikaël Salson committed
355 356 357 358 359 360 361 362
}

template <class T>
MapKmerStore<T>::MapKmerStore(int k, bool revcomp){
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
363 364 365 366 367 368
  init();
}

template <class T>
void MapKmerStore<T>::init() {
  this->nb_kmers_inserted = 0;
Mikaël Salson's avatar
Mikaël Salson committed
369 370 371
}

template <class T> 
372
T& MapKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
373 374 375 376
  return store[word];
}

template <class T> 
377
T& MapKmerStore<T>::operator[](seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
378
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
379
    seqtype rc_kmer = revcomp(word);
Mikaël Salson's avatar
Mikaël Salson committed
380 381 382 383 384 385 386 387 388
    if (rc_kmer.compare(word) < 0)
      word = rc_kmer;
  }
  return store[word];
}

// ArrayKmerStore

template <class T> 
389
ArrayKmerStore<T>::ArrayKmerStore(int k, bool revcomp) {
Mikaël Salson's avatar
Mikaël Salson committed
390 391 392 393
  this->seed = seed_contiguous(k);
  this->k = k;
  this->s = k;
  this->revcomp_indexed = revcomp;
394
  init();
Mikaël Salson's avatar
Mikaël Salson committed
395 396 397 398 399 400 401 402 403 404
}


template <class T> 
ArrayKmerStore<T>::ArrayKmerStore(string seed, bool revcomp){
  this->seed = seed; 
  int k = seed_weight(seed);
  this->k = k;
  this->s = seed.size();
  this->revcomp_indexed = revcomp;
405 406 407 408 409
  init();
}

template <class T>
void ArrayKmerStore<T>::init() {
410
  this->nb_kmers_inserted = 0;
411
  if ((size_t)(this->k << 1) >= sizeof(int) * 8)
Mikaël Salson's avatar
Mikaël Salson committed
412
    throw std::bad_alloc();
413
  store = new T[(unsigned int)1 << (this->k << 1)];
Mikaël Salson's avatar
Mikaël Salson committed
414 415 416 417 418 419 420 421 422 423 424 425 426 427
  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> 
428
int ArrayKmerStore<T>::index(const seqtype& word) const{
Mikaël Salson's avatar
Mikaël Salson committed
429
  return dna_to_int(word, this->k);
Mikaël Salson's avatar
Mikaël Salson committed
430 431 432
}

template <class T> 
433
T& ArrayKmerStore<T>::get(seqtype& word){
Mikaël Salson's avatar
Mikaël Salson committed
434 435 436 437
  return store[index(word)];
}

template <class T> 
438
T& ArrayKmerStore<T>::operator[](seqtype& word){
439 440 441 442 443
  return (*this)[index(word)];
}

template <class T> 
T& ArrayKmerStore<T>::operator[](int word){
Mikaël Salson's avatar
Mikaël Salson committed
444
  if (this->isRevcomp() && T::hasRevcompSymetry()) {
445 446
    int rc_kmer = revcomp_int(word, IKmerStore<T>::k);
    if (rc_kmer < word)
Mikaël Salson's avatar
Mikaël Salson committed
447 448
      word = rc_kmer;
  }
449
  return store[word];
Mikaël Salson's avatar
Mikaël Salson committed
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470
}


/**
 * 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){
471
    cout << "  (using a MapKmer to fit into memory)" << endl;
Mikaël Salson's avatar
Mikaël Salson committed
472 473 474 475 476 477 478 479 480 481 482
    index = new MapKmerStore<T>(seed, revcomp);
  }
  return index;
}

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

#endif