-
Notifications
You must be signed in to change notification settings - Fork 53
/
Data.cpp
executable file
·2644 lines (2030 loc) · 95.1 KB
/
Data.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
This file is part of the regenie software package.
Copyright (c) 2020 Joelle Mbatchou & Jonathan Marchini
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
*/
#include <limits.h> /* for PATH_MAX */
#include <boost/filesystem.hpp>
#include "Regenie.hpp"
#include "Geno.hpp"
#include "Step1_Models.hpp"
#include "Step2_Models.hpp"
#include "Files.hpp"
#include "Pheno.hpp"
#include "Data.hpp"
using namespace std;
using namespace Eigen;
using namespace boost;
namespace fs = boost::filesystem;
template <typename T> int sgn(T val) {
return (T(0) < val) - (val < T(0));
}
using namespace Eigen;
using boost::math::normal;
using boost::math::chi_squared;
Data::Data() { // @suppress("Class members should be properly initialized")
}
Data::~Data() {
// TODO Auto-generated destructor stub
}
void Data::run() {
if(params.test_mode) {
if(params.streamBGEN) check_bgen(files.bgen_file, ¶ms);
if(params.streamBGEN) test_snps_fast();
else test_snps();
} else {
sout << "Fitting null model\n";
// set number of threads
setNbThreads(params.threads);
// set up file for reading
file_read_initialization();
// read phenotype and covariate files
read_pheno_and_cov(&files, ¶ms, &in_filters, &pheno_data, &m_ests, sout);
// adjust for covariates
prep_run(&files, ¶ms, &pheno_data, &m_ests, sout);
// set number of blocks and block size and ridge parameters
set_blocks();
// some initializations
setmem();
// level 0
level_0_calculations();
// level 1 ridge
if(!params.binary_mode) // QT
if(params.use_loocv) ridge_level_1_loocv(&files, ¶ms, &pheno_data, &l1_ests, sout);
else ridge_level_1(&files, ¶ms, &l1_ests, sout);
else // BT
if(params.use_loocv) ridge_logistic_level_1_loocv(&files, ¶ms, &pheno_data, &m_ests, &l1_ests, sout);
else ridge_logistic_level_1(&files, ¶ms, &pheno_data, &l1_ests, masked_in_folds, sout);
// output results
output();
}
}
/////////////////////////////////////////////////
/////////////////////////////////////////////////
//// read in files
/////////////////////////////////////////////////
/////////////////////////////////////////////////
void Data::file_read_initialization() {
// prepare genotype data
files.chr_counts.assign(params.nChrom, 0.0);
if(params.file_type == "bed") read_bed_bim_fam(&files, ¶ms, &in_filters, snpinfo, chr_map, sout);
else if(params.file_type == "pgen") read_pgen_pvar_psam(&files, ¶ms, &in_filters, &Gblock, snpinfo, chr_map, sout);
else prep_bgen(&files, ¶ms, &in_filters, snpinfo, chr_map, Gblock.bgen, sout);
if( params.setMinINFO && !params.dosage_mode )
sout << "WARNING: Dosages are not present in the genotype file. Option --minINFO is skipped.\n";
}
/////////////////////////////////////////////////
/////////////////////////////////////////////////
//// adjust for covariates in G
/////////////////////////////////////////////////
/////////////////////////////////////////////////
void Data::residualize_genotypes() {
sout << " -residualizing and scaling genotypes..." << flush;
auto t1 = std::chrono::high_resolution_clock::now();
// mask missing individuals
Gblock.Gmat.array().rowwise() *= in_filters.ind_in_analysis.matrix().transpose().array().cast<double>();
if(params.strict_mode) Gblock.Gmat.array().rowwise() *= pheno_data.masked_indivs.col(0).matrix().transpose().array().cast<double>();
// residuals (centered)
MatrixXd beta = Gblock.Gmat * pheno_data.new_cov;
Gblock.Gmat -= beta * pheno_data.new_cov.transpose();
// scaling (use [N-C] where C=#covariates)
if(params.strict_mode) scale_G = Gblock.Gmat.rowwise().norm() / sqrt(pheno_data.Neff(0) - params.ncov);
else scale_G = Gblock.Gmat.rowwise().norm() / sqrt(in_filters.ind_in_analysis.cast<double>().sum() - params.ncov);
// check sd
MatrixXd::Index minIndex;
if(scale_G.array().minCoeff(&minIndex) < params.numtol) {
if(!params.test_mode) {
sout << "!! Uh-oh, SNP " << snpinfo[in_filters.step1_snp_count+minIndex].ID << " has low variance (=" << scale_G(minIndex,0) << ").\n";
exit( EXIT_FAILURE );
} else {
Gblock.bad_snps = scale_G.array() < params.numtol;
for(int i = 0; i < Gblock.Gmat.rows(); i++){
// make snps polymorphic
if(Gblock.bad_snps(i)) {
Gblock.Gmat.row(i).head(20).array() += 1; // +1 to first 20 entries
scale_G(i) = 1;
if(params.verbose) sout << "WARNING: Ignoring SNP with low variance.\n";
}
}
}
}
Gblock.Gmat.array().colwise() /= scale_G.array();
// to use MAF dependent prior on effect size [only for step 1]
// multiply by [p*(1-p)]^(1+alpha)/2
if( !params.test_mode && (params.alpha_prior != -1) ){
Gblock.Gmat.array().colwise() *= pow(Gblock.snp_afs.col(0).array() * (1-Gblock.snp_afs.col(0).array()), 0.5 * (params.alpha_prior + 1) );
}
sout << "done";
auto t2 = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
sout << " (" << duration.count() << "ms) "<< endl;
}
/////////////////////////////////////////////////
/////////////////////////////////////////////////
//// step 1: prepare for level 0
/////////////////////////////////////////////////
/////////////////////////////////////////////////
void Data::set_blocks() {
params.total_n_block = 0, total_chrs_loco = 0;
int blocks_left = params.n_block;
uint32_t n_analyzed = in_filters.ind_in_analysis.cast<int>().sum();
map<int, vector<int> >::iterator itr;
map<int, vector<int> > m1;
// compute number of blocks for each chromosome
for (itr = chr_map.begin(); itr != chr_map.end(); ++itr) {
int chrom_nsnps = itr->second[0];
int nb = ceil(chrom_nsnps * 1.0 / params.block_size);
if(params.n_block > 0) {
if(blocks_left > 0) {
int minb = min(nb, blocks_left);
//sout << << endl;
itr->second[1] = minb;
params.total_n_block += minb;
blocks_left -= minb;
}
} else {
itr->second[1] = nb;
params.total_n_block += nb;
}
// track how many chromosome will have blups
if(itr->second[1] > 0) total_chrs_loco++;
m1.insert(pair<int, vector<int> >(itr->first, itr->second));
}
chr_map = m1;
//sout << "#chrs = "<< chr_map.size() << ";#loco chrs = "<< total_chrs_loco << endl;
if(params.total_n_block == 0){
sout << "ERROR: Total number of blocks must be > 0.\n";
exit( EXIT_FAILURE );
}
// set ridge params
for(int i = 0; i < params.n_ridge_l0; i++) params.lambda[i] = params.n_variants / params.lambda[i];
for(int i = 0; i < params.n_ridge_l1; i++) {
if(!params.binary_mode)
params.tau[i] = (params.total_n_block * params.n_ridge_l0) / params.tau[i];
else {
// Assuming input tau[i] is total SNP heritability on the liability scale= m * 3/pi^2 * (1-h2) / h2
params.tau[i] = (params.total_n_block * params.n_ridge_l0) * 3 / (M_PI * M_PI) * (1 - params.tau[i]) / params.tau[i];
}
}
// for BTs: check if the sample size is lower than 5K (if so, use loocv)
if( params.binary_mode && ( n_analyzed < 5000) ) {
if(!params.use_loocv){
sout << " -WARNING: Sample size is less than 5,000 so using LOOCV instead of " << params.cv_folds << "-fold CV.\n";
params.use_loocv = true;
}
}
/*
// check block size vs sample size
if(params.use_loocv && params.block_size > n_analyzed){
sout << "ERROR: Block size must be smaller than the number of samples to perform LOOCV!\n";
exit( EXIT_FAILURE );
}
*/
if(params.use_loocv) params.cv_folds = params.n_samples;
uint32_t neff_folds = params.use_loocv ? n_analyzed : params.cv_folds;
// summarize block sizes and ridge params
sout << left << std::setw(20) << " * # threads" << ": [" << params.threads << "]\n";
sout << left << std::setw(20) << " * block size" << ": [" << params.block_size << "]\n";
sout << left << std::setw(20) << " * # blocks" << ": [" << params.total_n_block << "]\n";
sout << left << std::setw(20) << " * # CV folds" << ": [" << neff_folds << "]\n";
sout << left << std::setw(20) << " * ridge data_l0" << ": [" << params.n_ridge_l0 << " : ";
for(int i = 0; i < params.n_ridge_l0; i++) sout << params.n_variants / params.lambda[i] << " ";
sout << "]\n";
sout << left << std::setw(20) << " * ridge data_l1" << ": [" << params.n_ridge_l1 << " : ";
for(int i = 0; i < params.n_ridge_l1; i++) {
if(!params.binary_mode)
sout << (params.total_n_block * params.n_ridge_l0) / params.tau[i] << " ";
else
sout << (params.total_n_block * params.n_ridge_l0) / ( (params.total_n_block * params.n_ridge_l0) + (M_PI * M_PI) * params.tau[i] / 3 ) << " ";
}
sout << "]\n";
// if using maf dependent prior
if(!params.test_mode && (params.alpha_prior != -1) ) sout << " * applying a MAF dependent prior to the SNP effect sizes in level 0 models (alpha=" << params.alpha_prior << ")\n";
// print approx. amount of memory needed
print_usage_info(¶ms, &files, sout);
// if within sample predictions are used in level 1
if (params.within_sample_l0) {
sout << " * using within-sample predictions from level 0 as features at level 1\n";
}
}
void Data::set_folds() {
// set up folds
params.cv_sizes.resize(params.cv_folds);
if(params.strict_mode && !params.use_loocv){
// assign folds accounting for missingness
uint32_t target_size_folds = floor( pheno_data.Neff(0) / params.cv_folds );
if( target_size_folds < 1 ){
sout << "ERROR: Not enough samples are present for " << params.cv_folds<<"-fold CV.\n";
exit( EXIT_FAILURE );
}
uint32_t n_non_miss = 0, cum_size_folds = 0;
int cur_fold = 0;
for(size_t i = 0; i < params.n_samples; i++){
if( pheno_data.masked_indivs(i, 0) ) n_non_miss++;
if( n_non_miss == target_size_folds){
params.cv_sizes[cur_fold] = i - cum_size_folds + 1;
cum_size_folds += params.cv_sizes[cur_fold];
n_non_miss = 0, cur_fold++;
} else if( cur_fold == (params.cv_folds - 1) ){
params.cv_sizes[cur_fold] = params.n_samples - i;
break;
}
//sout << i << " " << cur_fold << " " << n_non_miss << " " << (pheno_data.masked_indivs(i, 0)?"Y":"N") << " "<< target_size_folds << endl;
}
} else {
// assign folds for individuals in analysis
if( !params.use_loocv ){
uint32_t target_size_folds = floor( in_filters.ind_in_analysis.cast<double>().sum() / params.cv_folds );
if( target_size_folds < 1 ){
sout << "ERROR: Not enough samples are present for " << params.cv_folds<<"-fold CV.\n";
exit( EXIT_FAILURE );
}
uint32_t n_non_miss = 0, cum_size_folds = 0;
int cur_fold = 0;
for(size_t i = 0; i < params.n_samples; i++){
if( in_filters.ind_in_analysis(i) ) n_non_miss++;
if( n_non_miss == target_size_folds){
params.cv_sizes[cur_fold] = i - cum_size_folds + 1;
cum_size_folds += params.cv_sizes[cur_fold];
n_non_miss = 0, cur_fold++;
} else if( cur_fold == (params.cv_folds - 1) ){
params.cv_sizes[cur_fold] = params.n_samples - i;
break;
}
//sout << i << " " << cur_fold << " " << n_non_miss << " " << in_filters.ind_in_analysis(i) << " "<< target_size_folds << endl;
}
} else { // loocv
for(int i = 0; i < params.cv_folds; i++) params.cv_sizes[i] = 1;
}
}
// check sd(Y) in folds
if(!params.use_loocv && params.binary_mode){
uint32_t cum_size_folds = 0;
MatrixXd phenos = ( pheno_data.phenotypes_raw.array() * pheno_data.masked_indivs.array().cast<double>()).matrix();
for(int i = 0; i < (params.cv_folds - 1); i++) {
ArrayXd sum = phenos.block(cum_size_folds,0,params.cv_sizes[i],params.n_pheno).colwise().sum();
ArrayXd n_cv = pheno_data.masked_indivs.block(cum_size_folds,0,params.cv_sizes[i],params.n_pheno).cast<double>().colwise().sum();
ArrayXd sd_phenos = (sum/n_cv) * (1 - sum/n_cv);
if( sd_phenos.minCoeff() < params.numtol ){
sout << "ERROR: One of the folds has only cases/controls! Either use smaller #folds (option --cv) or use LOOCV (option --loocv).\n";
exit( EXIT_FAILURE );
}
cum_size_folds += params.cv_sizes[i];
}
}
// only used for K-fold CV
if(!params.use_loocv && !params.within_sample_l0){
l1_ests.X_folds.resize(params.cv_folds);
l1_ests.XtY.resize(params.cv_folds);
}
}
void Data::setmem() {
sout << " * setting memory..." << flush;
Gblock.Gmat = MatrixXd::Zero(params.block_size, params.n_samples);
set_folds();
l1_ests.cumsum_values.resize(6);
predictions.resize(1);
predictions[0] = MatrixXd::Zero(params.n_samples, total_chrs_loco);
if (params.within_sample_l0) {
l1_ests.pred_mat.resize(params.n_pheno);
l1_ests.pred_pheno.resize(params.n_pheno);
} else if(!params.use_loocv) l1_ests.beta_hat_level_1.resize(params.n_pheno);
if(!params.use_loocv) {
l1_ests.test_pheno.resize(params.n_pheno);
l1_ests.test_mat.resize(params.n_pheno);
} else l1_ests.test_mat_conc.resize(params.n_pheno);
if(params.binary_mode){
if (params.within_sample_l0) {
l1_ests.pred_pheno_raw.resize(params.n_pheno);
l1_ests.pred_offset.resize(params.n_pheno);
}
l1_ests.test_pheno_raw.resize(params.n_pheno);
if(!params.use_loocv) l1_ests.test_offset.resize(params.n_pheno);
}
masked_in_folds.resize(params.cv_folds);
if(params.print_block_betas) params.beta_print_out.resize(params.n_pheno);
for(int i = 0; i < params.n_pheno; ++i ) {
if (params.within_sample_l0) {
l1_ests.pred_mat[i].resize(params.cv_folds);
l1_ests.pred_pheno[i].resize(params.cv_folds);
} else if(!params.use_loocv) l1_ests.beta_hat_level_1[i].resize(params.cv_folds);
if(!params.use_loocv) {
l1_ests.test_pheno[i].resize(params.cv_folds);
l1_ests.test_mat[i].resize(params.cv_folds);
} else l1_ests.test_mat_conc[i] = MatrixXd::Zero(params.n_samples, params.n_ridge_l0 * ( params.write_l0_pred ? 1 : params.total_n_block) );
if(params.binary_mode) {
if (params.within_sample_l0) {
l1_ests.pred_pheno_raw[i].resize(params.cv_folds);
l1_ests.pred_offset[i].resize(params.cv_folds);
}
l1_ests.test_pheno_raw[i].resize(params.cv_folds);
if(!params.use_loocv) l1_ests.test_offset[i].resize(params.cv_folds);
}
for(int j = 0; j < params.cv_folds; ++j ) {
if (params.within_sample_l0) {
l1_ests.pred_mat[i][j] = MatrixXd::Zero(params.n_samples - params.cv_sizes[j], params.total_n_block * params.n_ridge_l0);
l1_ests.pred_pheno[i][j] = MatrixXd::Zero(params.n_samples - params.cv_sizes[j], 1);
} else if(!params.use_loocv) l1_ests.beta_hat_level_1[i][j] = MatrixXd::Zero(params.total_n_block * params.n_ridge_l0, params.n_ridge_l1);
if(!params.use_loocv) {
l1_ests.test_pheno[i][j] = MatrixXd::Zero(params.cv_sizes[j], 1);
l1_ests.test_mat[i][j] = MatrixXd::Zero(params.cv_sizes[j], params.n_ridge_l0 * ( params.write_l0_pred ? 1 : params.total_n_block));
}
if(params.binary_mode) {
if (params.within_sample_l0) {
l1_ests.pred_pheno_raw[i][j] = MatrixXd::Zero(params.n_samples - params.cv_sizes[j], 1);
l1_ests.pred_offset[i][j] = MatrixXd::Zero(params.n_samples - params.cv_sizes[j], 1);
}
l1_ests.test_pheno_raw[i][j] = MatrixXd::Zero(params.cv_sizes[j], 1);
if(!params.use_loocv) l1_ests.test_offset[i][j] = MatrixXd::Zero(params.cv_sizes[j], 1);
}
if(i == 0) masked_in_folds[j] = MatrixXb::Constant(params.cv_sizes[j], params.n_pheno, false);
}
}
sout << "done\n" << endl;
}
/////////////////////////////////////////////////
/////////////////////////////////////////////////
//// step 1: level 0
/////////////////////////////////////////////////
/////////////////////////////////////////////////
void Data::level_0_calculations() {
int block = 0, nread=0, chrom_nsnps_file;
if(params.print_block_betas) params.print_snpcount = 0;
snp_index_counter = 0;
ridgel0 l0;
if(!params.use_loocv){
l0.G_folds.resize(params.cv_folds);
l0.GtY.resize(params.cv_folds);
}
for (size_t itr = 0; itr < files.chr_read.size(); ++itr) {
int chrom = files.chr_read[itr];
if( chr_map.find(chrom) == chr_map.end() ) continue;
int chrom_nsnps = chr_map[chrom][0];
int chrom_nb = chr_map[chrom][1];
if(params.keep_snps || params.rm_snps) chrom_nsnps_file = chr_map[chrom][2];
// if all snps in chromosome are excluded
// stream through file to read next chr
if((params.keep_snps || params.rm_snps) && (chrom_nb == 0)) {
skip_snps(chrom_nsnps_file, ¶ms, &files, &Gblock);
snp_index_counter += chrom_nsnps_file;
continue;
}
if(chrom_nb > 0) sout << "Chromosome " << chrom << endl;
//sout << "Ns="<< chrom_nsnps << ";Nfile="<< chrom_nsnps_file << endl;
if(params.keep_snps || params.rm_snps) nread = snp_index_counter;
for(int bb = 0; bb < chrom_nb ; bb++) {
int bs = params.block_size;
if(bb == 0) {
Gblock.Gmat = MatrixXd::Zero(bs, params.n_samples);
if(params.alpha_prior != -1) Gblock.snp_afs = MatrixXd::Zero(bs, 1);
}
if((bb + 1) * params.block_size > chrom_nsnps) {
bs = chrom_nsnps - (bb * params.block_size) ;
Gblock.Gmat = MatrixXd::Zero(bs, params.n_samples);
if(params.alpha_prior != -1) Gblock.snp_afs = MatrixXd::Zero(bs, 1);
}
get_G(block, bs, chrom, snp_index_counter, snpinfo, ¶ms, &files, &Gblock, &in_filters, pheno_data.masked_indivs, pheno_data.phenotypes_raw, sout);
// residualize and scale genotypes
residualize_genotypes();
// calc working matrices for ridge regressions across folds
calc_cv_matrices(bs, &l0);
// calc level 0 ridge regressions
if(params.use_loocv)
ridge_level_0_loocv(block, &files, ¶ms, &in_filters, &m_ests, &Gblock, &pheno_data, snpinfo, &l0, &l1_ests, sout);
else
ridge_level_0(block, &files, ¶ms, &in_filters, &m_ests, &Gblock, &pheno_data, snpinfo, &l0, &l1_ests, masked_in_folds, sout);
block++; in_filters.step1_snp_count += bs;
}
// if skipping all snps at end of chromosome
if(params.keep_snps || params.rm_snps){
nread = snp_index_counter - nread;
//sout << "Nread =" << nread << "; chr=" << chrom << endl;
if(nread < chrom_nsnps_file) {
skip_snps(chrom_nsnps_file - nread, ¶ms, &files, &Gblock);
snp_index_counter += chrom_nsnps_file - nread;
//sout << "Nskipping=" << chrom_nsnps_file - nread << endl;
}
}
}
if(params.early_exit) {
sout << "\nDone printing out level 0 predictions. There are " <<
params.n_samples << " rows and " <<
params.total_n_block * params.n_ridge_l0 << " columns " <<
"stored in column-major order. Exiting...\n";
runtime.stop();
sout << "\nElapsed time : " << std::chrono::duration<double>(runtime.end - runtime.begin).count() << "s\n";
sout << "End time: " << ctime(&runtime.end_time_info) << endl;
exit( EXIT_SUCCESS );
}
// free up memory not used anymore
Gblock.Gmat.resize(0,0);
if(params.write_l0_pred && (params.n_pheno > 1) ) {
// free level 0 predictions for (P-1) indices in test_mat
for(int ph = 1; ph < params.n_pheno; ++ph ) {
if(!params.use_loocv){
for(int i = 0; i < params.cv_folds; ++i ) l1_ests.test_mat[ph][i].resize(0,0);
l1_ests.test_mat[ph].resize(0);
} else {
l1_ests.test_mat_conc[ph].resize(0,0);
}
}
}
}
void Data::calc_cv_matrices(const int bs, struct ridgel0* l0) {
sout << " -calc working matrices..." << flush;
auto t2 = std::chrono::high_resolution_clock::now();
if(!params.use_loocv){
l0->GGt.setZero(bs,bs);
l0->GTY.setZero(bs,params.n_pheno);
uint32_t cum_size_folds = 0;
for( int i = 0; i < params.cv_folds; ++i ) {
l0->GtY[i] = Gblock.Gmat.block(0, cum_size_folds, bs, params.cv_sizes[i]) * pheno_data.phenotypes.block(cum_size_folds, 0, params.cv_sizes[i], params.n_pheno);
l0->G_folds[i] = Gblock.Gmat.block(0, cum_size_folds, bs, params.cv_sizes[i]) * Gblock.Gmat.block(0, cum_size_folds, bs, params.cv_sizes[i]).transpose();
l0->GGt += l0->G_folds[i];
l0->GTY += l0->GtY[i];
cum_size_folds += params.cv_sizes[i];
}
} else {
l0->GGt = Gblock.Gmat * Gblock.Gmat.transpose();
l0->GTY = Gblock.Gmat * pheno_data.phenotypes;
SelfAdjointEigenSolver<MatrixXd> esG(l0->GGt);
l0->GGt_eig_vec = esG.eigenvectors();
l0->GGt_eig_val = esG.eigenvalues();
l0->Wmat = l0->GGt_eig_vec.transpose() * l0->GTY;
}
sout << "done";
auto t3 = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(t3 - t2);
sout << " (" << duration.count() << "ms) "<< endl;
}
/////////////////////////////////////////////////
/////////////////////////////////////////////////
//// Evaluate level 1 output
/////////////////////////////////////////////////
/////////////////////////////////////////////////
void Data::output() {
int min_index;
double performance_measure, rsq, sse, ll_avg, min_val;
string pfile, out_blup_list, out_prs_list, loco_filename, prs_filename;
string fullpath_str, path_prs;
Files outb, outp;
sout << "Output\n" << "------\n";
if(params.make_loco || params.binary_mode){
out_blup_list = files.out_file + "_pred.list";
outb.openForWrite(out_blup_list, sout);
}
if(params.print_prs){
out_prs_list = files.out_file + "_prs.list";
outp.openForWrite(out_prs_list, sout);
}
for(int ph = 0; ph < params.n_pheno; ++ph ) {
sout << "phenotype " << ph+1 << " (" << files.pheno_names[ph] << ") : " ;
loco_filename = files.out_file + "_" + to_string(ph + 1) + ".loco" + (params.gzOut ? ".gz" : "");
prs_filename = files.out_file + "_" + to_string(ph + 1) + ".prs" + (params.gzOut ? ".gz" : "");
if( params.make_loco || params.binary_mode || params.print_prs ) {
fullpath_str = get_fullpath(loco_filename);
if(params.print_prs) path_prs = get_fullpath(prs_filename);
if( !params.binary_mode ) { // for quantitative traits
outb << files.pheno_names[ph] << " " << fullpath_str << endl;
if(params.print_prs) outp << files.pheno_names[ph] << " " << path_prs << endl;
} else { // for binary traits - check level 1 ridge converged
if( !l1_ests.pheno_l1_not_converged(ph) ) {
outb << files.pheno_names[ph] << " " << fullpath_str << endl;
if(params.print_prs) outp << files.pheno_names[ph] << " " << path_prs << endl;
} else {
if(params.write_l0_pred){ // cleanup level 0 predictions
pfile = files.loco_tmp_prefix + "_l0_Y" + to_string(ph+1);
remove(pfile.c_str());
}
sout << "Level 1 logistic did not converge. LOCO predictions calculations are skipped.\n\n";
continue;
}
}
}
sout << endl;
min_index = 0;
min_val = 1e10;
// determine optimal parameter by cv using: QT: MSE, BT: -loglik
for(int j = 0; j < params.n_ridge_l1; ++j ) {
if(!params.binary_mode)
performance_measure = l1_ests.cumsum_values[2](ph, j) + l1_ests.cumsum_values[3](ph,j) - 2 * l1_ests.cumsum_values[4](ph,j);
else
performance_measure = l1_ests.cumsum_values[5](ph, j);
performance_measure /= pheno_data.Neff(ph);
if( performance_measure < min_val) {
min_index = j;
min_val = performance_measure;
}
}
for(int j = 0; j < params.n_ridge_l1; ++j ) {
if(!params.binary_mode)
sout << " " << setw(5) << (params.total_n_block * params.n_ridge_l0) / params.tau[j] ;
else
sout << " " << setw(5) << (params.total_n_block * params.n_ridge_l0) / ( (params.total_n_block * params.n_ridge_l0) + (M_PI * M_PI) * params.tau[j] / 3 );
// output Rsq and MSE
rsq = l1_ests.cumsum_values[4](ph,j) - l1_ests.cumsum_values[0](ph,j) * l1_ests.cumsum_values[1](ph,j) / pheno_data.Neff(ph); // num = Sxy - SxSy/n
rsq = (rsq * rsq) / ((l1_ests.cumsum_values[2](ph,j) - l1_ests.cumsum_values[0](ph,j) * l1_ests.cumsum_values[0](ph,j) / pheno_data.Neff(ph)) * (l1_ests.cumsum_values[3](ph,j) - l1_ests.cumsum_values[1](ph,j) * l1_ests.cumsum_values[1](ph,j) / pheno_data.Neff(ph))); // num^2 / ( (Sx2 - Sx^2/n)* (Sy2 - Sy^2/n) )
sse = l1_ests.cumsum_values[2](ph, j) + l1_ests.cumsum_values[3](ph,j) - 2 * l1_ests.cumsum_values[4](ph,j); // Sx2 + Sy2 - SxSy
if(params.binary_mode) ll_avg = l1_ests.cumsum_values[5](ph, j) / pheno_data.Neff(ph);
sout << " : ";
sout << "Rsq = " << rsq << ", MSE = " << sse/pheno_data.Neff(ph);
if(params.binary_mode) sout << ", -logLik/N = " << ll_avg;
if(j == min_index) sout << "<- min value";
sout << endl;
}
if(!params.binary_mode){
if(params.use_loocv) make_predictions_loocv(ph, min_index);
else make_predictions(ph, min_index);
} else if(params.use_loocv) make_predictions_binary_loocv(ph, min_index);
else make_predictions_binary(ph, min_index);
// delete file used to store l0 predictions
if(params.write_l0_pred){
pfile = files.loco_tmp_prefix + "_l0_Y" + to_string(ph+1);
remove(pfile.c_str());
}
}
if(params.make_loco || params.binary_mode){
outb.closeFile();
sout << "List of blup files written to: [" << out_blup_list << "]\n";
}
if(params.print_prs) {
outp.closeFile();
sout << "List of files with whole genome PRS written to: [" <<
out_prs_list << "]\n";
}
}
// convert filename to full path
std::string get_fullpath(std::string fname){
string fout;
try {
// convert to full path using boost filesystem library
// this can generate errors due to LC_ALL locale being invalid
fs::path fullpath;
fullpath = fs::absolute(fname);
fout = fullpath.make_preferred().string();
} catch ( std::runtime_error& ex ) {
try {
// use realpath
char buf[PATH_MAX];
char *res = realpath(fname.c_str(), buf);
if(res) fout = string(buf);
else fout = fname; // if failed to get full path
} catch ( const std::bad_alloc& ) {
fout = fname; // if failed to get full path
}
}
return fout;
}
void Data::make_predictions(const int ph, const int val) {
sout << " * making predictions..." << flush;
auto t1 = std::chrono::high_resolution_clock::now();
int bs_l1 = params.total_n_block * params.n_ridge_l0;
int ph_eff = params.write_l0_pred ? 0 : ph;
string outname, in_pheno;
ifstream infile;
ofstream ofile;
MatrixXd X1, X2, beta_l1, beta_avg;
MatrixXd ident_l1 = MatrixXd::Identity(bs_l1,bs_l1);
// read in level 0 predictions from file
if(params.write_l0_pred){
in_pheno = files.loco_tmp_prefix + "_l0_Y" + to_string(ph+1);
infile.open(in_pheno.c_str(), ios::in | ios::binary );
if (!infile.is_open()) {
sout << "ERROR : Cannot read temporary file " << in_pheno << endl ;
exit(EXIT_FAILURE);
}
// store back values in test_mat
for( int m = 0; m < bs_l1; ++m )
for( int i = 0; i < params.cv_folds; ++i )
for( int k = 0; k < params.cv_sizes[i]; ++k )
infile.read( reinterpret_cast<char *> (&l1_ests.test_mat[ph_eff][i](k,m)), sizeof(double) );
infile.close();
}
if(params.within_sample_l0){
X1 = l1_ests.test_mat[ph_eff][0].transpose() * l1_ests.test_mat[ph_eff][0];
X2 = l1_ests.test_mat[ph_eff][0].transpose() * l1_ests.test_pheno[ph][0];
for(int i = 1; i < params.cv_folds; ++i ) {
X1 += l1_ests.test_mat[ph_eff][i].transpose() * l1_ests.test_mat[ph_eff][i];
X2 += l1_ests.test_mat[ph_eff][i].transpose() * l1_ests.test_pheno[ph][i];
}
beta_l1 = (X1 + params.tau[val] * ident_l1).llt().solve(X2);
} else if(params.print_block_betas) {
beta_avg = MatrixXd::Zero(bs_l1, 1);
for(int i = 0; i < params.cv_folds; ++i ) {
beta_avg += l1_ests.beta_hat_level_1[ph][i].col(val);
}
beta_avg /= params.cv_folds;
}
// if specified, write betas to file (open in append mode)
if(!params.within_sample_l0 && params.print_block_betas) {
outname = files.out_file + "_level1.betas";
ofile.open(outname.c_str(), ios::out | ios::app);
ofile << ph + 1 << " ";
ofile << beta_avg.transpose() << endl;
ofile.close();
}
// sout << "\nFor tau[" << val <<"] = " << params.tau[val] << endl << beta_l1 << endl ;
int ctr = 0, chr_ctr = 0;
int nn, cum_size_folds;
for (size_t itr = 0; itr < files.chr_read.size(); ++itr) {
int chrom = files.chr_read[itr];
if( chr_map.find(chrom) == chr_map.end() ) continue;
nn = chr_map[chrom][1] * params.n_ridge_l0;
if(nn > 0) {
cum_size_folds = 0;
for(int i = 0; i < params.cv_folds; ++i ) {
if(!params.within_sample_l0) beta_l1 = l1_ests.beta_hat_level_1[ph][i].col(val);
predictions[0].block(cum_size_folds, chr_ctr, params.cv_sizes[i], 1) = l1_ests.test_mat[ph_eff][i].block(0, ctr, params.cv_sizes[i], nn) * beta_l1.block(ctr, 0, nn, 1);
cum_size_folds += params.cv_sizes[i];
}
chr_ctr++;
ctr += nn;
}
}
write_predictions(ph);
sout << "done";
auto t2 = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
sout << " (" << duration.count() << "ms) "<< endl << endl;
}
void Data::make_predictions_loocv(const int ph, const int val) {
sout << " * making predictions..." << flush;
auto t1 = std::chrono::high_resolution_clock::now();
int bs_l1 = params.total_n_block * params.n_ridge_l0;
int ph_eff = params.write_l0_pred ? 0 : ph;
string in_pheno;
ifstream infile;
MatrixXd Xmat_chunk, Yvec_chunk, Z1, Z2, b0, xtx;
VectorXd w1, w2, Vw2, zvec;
RowVectorXd calFactor;
ArrayXd dl_inv;
uint64 max_bytes = params.chunk_mb * 1e6;
// amount of RAM used < max_mb [ creating (target_size * bs_l1) matrix ]
int nchunk = ceil( params.cv_folds * bs_l1 * sizeof(double) * 1.0 / max_bytes );
if (params.verbose) sout << nchunk << " chunks..." << flush;
int chunk, size_chunk, target_size = params.cv_folds / nchunk;
int j_start;
// read in level 0 predictions from file
if(params.write_l0_pred){
in_pheno = files.loco_tmp_prefix + "_l0_Y" + to_string(ph+1);
infile.open(in_pheno.c_str(), ios::in | ios::binary );
if (!infile.is_open()) {
sout << "ERROR : Cannot read temporary file " << in_pheno << endl ;
exit(EXIT_FAILURE);
}
// store back values in test_mat_conc
infile.read( reinterpret_cast<char *> (&l1_ests.test_mat_conc[ph_eff](0,0)), params.n_samples * bs_l1 * sizeof(double) );
infile.close();
}
// fit model on whole data again for optimal ridge param
xtx = l1_ests.test_mat_conc[ph_eff].transpose() * l1_ests.test_mat_conc[ph_eff];
SelfAdjointEigenSolver<MatrixXd> eigX(xtx);
zvec = l1_ests.test_mat_conc[ph_eff].transpose() * pheno_data.phenotypes.col(ph);
w1 = eigX.eigenvectors().transpose() * zvec;
dl_inv = (eigX.eigenvalues().array() + params.tau[val]).inverse();
w2 = (w1.array() * dl_inv).matrix();
Vw2 = eigX.eigenvectors() * w2;
for(chunk = 0; chunk < nchunk; ++chunk ) {
size_chunk = chunk == nchunk - 1? params.cv_folds - target_size * chunk : target_size;
j_start = chunk * target_size;
Xmat_chunk = l1_ests.test_mat_conc[ph_eff].block(j_start, 0, size_chunk, bs_l1);
Yvec_chunk = pheno_data.phenotypes.block(j_start, ph, size_chunk, 1);
Z1 = (Xmat_chunk * eigX.eigenvectors()).transpose();
Z2 = dl_inv.matrix().asDiagonal() * Z1;
calFactor = (Z1.array() * Z2.array()).matrix().colwise().sum();
b0 = eigX.eigenvectors() * Z2;
b0.array().rowwise() *= (w2.transpose() * Z1 - Yvec_chunk.transpose()).array() / (1 - calFactor.array());
b0.colwise() += Vw2;
int ctr = 0, chr_ctr = 0;
int nn;
for (size_t itr = 0; itr < files.chr_read.size(); ++itr) {
int chrom = files.chr_read[itr];
if( chr_map.find(chrom) == chr_map.end() ) continue;
nn = chr_map[chrom][1] * params.n_ridge_l0;
if(nn > 0) {
predictions[0].block(j_start, chr_ctr, size_chunk, 1) = (l1_ests.test_mat_conc[ph_eff].block(j_start, ctr, size_chunk, nn).array() * b0.block(ctr, 0, nn, size_chunk).transpose().array()).rowwise().sum();
chr_ctr++;
ctr += nn;
}
}
}
write_predictions(ph);
sout << "done";
auto t2 = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
sout << " (" << duration.count() << "ms) "<< endl << endl;
}
// predictions for binary traits
void Data::make_predictions_binary(const int ph, const int val) {
sout << " * making predictions..." << flush;
auto t1 = std::chrono::high_resolution_clock::now();
int bs_l1 = params.total_n_block * params.n_ridge_l0;
int ph_eff = params.write_l0_pred ? 0 : ph;
string in_pheno;
ifstream infile;
ArrayXd etavec, pivec, wvec, zvec, score;
MatrixXd betaold, betanew, XtW, XtWX, XtWZ, Xconc;
MatrixXd ident_l1 = MatrixXd::Identity(bs_l1,bs_l1);
// read in level 0 predictions from file
if(params.write_l0_pred){
in_pheno = files.loco_tmp_prefix + "_l0_Y" + to_string(ph+1);
infile.open(in_pheno.c_str(), ios::in | ios::binary );
if (!infile.is_open()) {
sout << "ERROR : Cannot read temporary file " << in_pheno << endl ;
exit(EXIT_FAILURE);
}
// store back values in test_mat
for( int m = 0; m < bs_l1; ++m )
for( int i = 0; i < params.cv_folds; ++i )
for( int k = 0; k < params.cv_sizes[i]; ++k )
infile.read( reinterpret_cast<char *> (&l1_ests.test_mat[ph_eff][i](k,m)), sizeof(double) );
infile.close();
}
// fit model using out-of-sample level 0 predictions from whole data
if(params.within_sample_l0){
betaold = MatrixXd::Zero(bs_l1, 1);
int niter_cur = 0;
while(niter_cur++ < params.niter_max_ridge){
XtWX = MatrixXd::Zero(bs_l1, bs_l1);
XtWZ = MatrixXd::Zero(bs_l1, 1);
for(int i = 0; i < params.cv_folds; ++i ) {
etavec = (l1_ests.test_offset[ph][i] + l1_ests.test_mat[ph_eff][i] * betaold).array();
pivec = 1 - 1/(etavec.exp() + 1);
wvec = pivec * (1 - pivec);
zvec = (etavec - l1_ests.test_offset[ph][i].array()) + (l1_ests.test_pheno_raw[ph][i].array() - pivec) / wvec;
XtW = l1_ests.test_mat[ph_eff][i].transpose() * wvec.matrix().asDiagonal();
XtWX += XtW * l1_ests.test_mat[ph_eff][i];
XtWZ += XtW * zvec.matrix();
}
betanew = (XtWX + params.tau[val] * ident_l1).llt().solve(XtWZ);
// compute score
score = ArrayXd::Zero(betanew.rows());
for(int i = 0; i < params.cv_folds; ++i ) {
etavec = (l1_ests.test_offset[ph][i] + l1_ests.test_mat[ph_eff][i] * betanew).array();
pivec = 1 - 1/(etavec.exp() + 1);
score += (l1_ests.test_mat[ph_eff][i].transpose() * (l1_ests.test_pheno_raw[ph][i].array() - pivec).matrix()).array();
}
score -= params.tau[val] * betanew.array();
// stopping criterion
if( score.abs().maxCoeff() < params.l1_ridge_eps) break;
betaold = betanew;
}
}
// compute predictor for each chr
int ctr = 0, chr_ctr = 0;