Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 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 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 | 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 126x 126x 126x 126x 126x 126x 126x 126x 126x 71284x 71284x 71284x 71284x 71284x 71284x 71284x 71284x 126x 105x 105x 105x 1x 1x 1x 1x 1x 126x 21x 21x 34x 1x 1x 34x 1x 1x 34x 19x 123x 123x 126x 123x 123x 71276x 123x 123x 123x 71219x 71219x 71219x 489435x 489435x 92x 92x 71272x 71272x 34x 71238x 71272x 71238x 71238x 71238x 71238x 71238x 71238x 71238x 71238x 160437x 160437x 158424x 158424x 86180x 86180x 86180x 71238x 71238x 34x 34x 34x 68x 68x 81125x 81125x 81116x 81116x 81116x 81116x 81116x 32x 34x 34x 34x 122x 122x 122x 122x 122x 122x 122x 71272x 167296x 167296x 167296x 167296x 167296x 167296x 167296x 71272x 122x 120x 120x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 167296x 167296x 167296x 167296x 167296x 1348x 1348x 122x 122x 122x 122x 122x 122x 122x 71272x 71272x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 1348x 71272x 122x 122x 122x 122x 19x 19x 37x 34x 81116x 81116x 81116x 34x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 140x 140x 136x 136x 136x 18x 18x 122x 122x 140x 140x 140x 140x 140x 140x 140x 72364x 72364x 359329x 359329x 359329x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 21936x 359329x 359329x 359329x 359329x 359329x 359329x 359329x 72364x 72364x 140x 140x 136x 136x 136x 136x 141x 141x 141x 141x 141x 20936x 20936x 20936x 20936x 20936x 20936x 20936x 20936x 20936x 20936x 141x 141x 141x 141x 141x 141x 141x 141x 136x 136x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 403x 403x 403x 403x 403x 403x 398x 71272x 71272x 71272x 71272x 71272x 71272x 403x 5x 281x 281x 281x 281x 281x 5x 403x 122x 122x 261x 261x 261x 261x 261x 261x 261x 261x 2228x 2228x 2228x 2228x 2228x 2228x 2228x 105x 2228x 2228x 2228x 2228x 2228x 261x 261x 258x 258x 258x 258x 258x 258x 258x 258x 258x 258x 258x 2x 256x 258x 258x 258x 258x 258x 258x 258x 258x 258x 258x 258x 263x 263x 263x 263x 263x 263x 261x 1489x 1489x 1489x 1489x 1489x 1489x 1489x 1489x 1489x 263x 2x 5x 5x 5x 5x 5x 5x 5x 5x 5x 2x 263x 258x 258x 122x 122x 122x 122x 122x 122x 122x 122x 258x 258x 1489x 1489x 246x 246x 246x 1243x 1489x 1243x 1243x 258x 258x 122x 122x 125x 125x 125x 125x 122x 122x 122x 136x 136x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 2615x 2615x 2615x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 136x 136x 136x 136x 136x 136x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 125x 125x 125x 136x 136x 136x 136x 136x 125x 125x 125x 258x 258x 258x 258x 258x 258x 258x 258x 1489x 1489x 1489x 258x 258x 258x 122x 122x 122x 122x 122x 122x 122x 122x 122x 122x 258x 258x 258x 258x 258x 122x 122x 122x 122x 122x 2x 1x 1x 1x 1x 1x 4x 2021x 2021x 2021x 2021x 4x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x | /**
* A bigWig, written here, in this tab, with nothing installed.
*
* A bigWig is what a reader actually wants at the end of the day. It is what a
* genome browser opens, it is what a colleague expects to be sent, and it is the
* one file an analysis can produce that keeps working long after the app that
* made it is closed. Every alternative to writing it here ends in the same
* sentence — *now convert it with `bedGraphToBigWig`* — and that sentence is a
* command-line dependency in an application whose whole claim is that it has
* none (§18.7). So the writer is native, it adds no npm dependency, and it is
* read back through this repo's own `@gmod/bbi` in the test suite, because a
* writer nobody reads back is a writer that is wrong.
*
* **What makes it affordable is that the input is never arbitrary.** One value
* per bin, over an assembly whose contigs and lengths are already known, at a
* fixed bin size. There is no sorting to do, no interval arithmetic to get
* wrong, no variable-span sections to reason about, and a whole genome's worth
* of bins is kilobytes — so the file is assembled in memory and the header is
* written last, against offsets that are known by then. This is deliberately
* *not* a general BBI writer, and building one would have cost more and bought
* nothing this app can use.
*
* The six decisions of §18.7, each stated where it is taken below:
*
* 1. **bedGraph-typed sections (`type = 1`), never fixedStep.** Twelve bytes per
* item instead of four, which over a yeast genome is 145 kB against 48 kB
* before compression — irrelevant. What it buys is two correctness problems
* deleted rather than handled: a bin with no value is *omitted* instead of
* written as a `NaN` that every downstream tool would report as a value, and
* the last bin of a contig — partial, because a sequence length is not a
* multiple of the bin size — is written with its true end rather than
* overrunning the sequence. One code path and no clamping bug.
* 2. **zlib through the platform's own `CompressionStream`**, by way of
* `deflateBytes` in `ui/export.ts`, which already produces exactly the RFC
* 1950 framing BBI sections use for the PDF writer. Where it is absent the
* file goes out uncompressed with `uncompressBufSize = 0`, which is a legal
* bigWig and beats refusing to export.
* 3. **One bottom-up tree builder for both trees** — see `tree.ts`.
* 4. **Offsets by a push-and-running-length accumulator**, the same one
* `ui/export.ts` uses to place PDF objects.
* 5. **A zoom ladder from the base vector**: first reduction at ten times the bin
* size, then four times per level, ten levels at most, stopping at the level
* that holds fewer records than a tree node holds items. Every level is an
* aggregation of the base intervals, never a re-reading of a coarser one.
* 6. **Little-endian throughout**, which is what UCSC writes on x86 and what
* every reader detects from the magic number rather than being told.
*
* **Two header fields were settled by measurement rather than by reading a
* table**, since they are where writers most commonly disagree, and both were
* checked against the six real bigWigs in `demo/yeast/`:
*
* - The u64 at `fullDataOffset` is the number of **sections**, not the number of
* data items. In `Yeast_ChIP_Pol2-vs-input.bw` it reads 658, which is exactly
* the R-tree's leaf count, against 1,771,581 items actually on disk.
* - `itemsPerSlot` in an R-tree header is how many indexed things one leaf slot
* accounts for, which for the unzoomed index is **one section per slot** and
* therefore `1` — not the 1024 items a section happens to hold. The zoom
* indexes carry 1024 there, which is their records per block. Both are written
* that way below.
*
* Two further things the same measurement settled, because they are not
* obvious either: the total-summary `validCount` counts **bases covered** and
* not items, with `sumData` and `sumSquares` likewise weighted by each
* interval's span; and the count in front of a zoom level's data is a u32
* holding that level's **block** count, not its record count — the asymmetry
* with the u64 at `fullDataOffset` is real.
*/
import { deflateBytes } from '../../ui/export'
import { planTree, TREE_NODE_HEADER_BYTES } from './tree'
/** One sequence of the assembly. */
export interface BigWigContig {
name: string
length: number
}
/**
* One sequence's data as the intervals it actually is, for data on no grid.
*
* Ascending and non-overlapping, in the sequence's own coordinates. Values that
* are not finite are dropped, exactly as a `NaN` bin is on the other road — see
* {@link intervalsFor}.
*/
export interface BigWigSpans {
contig: string
starts: ArrayLike<number>
ends: ArrayLike<number>
values: ArrayLike<number>
}
/**
* Everything the writer needs, and nothing it does not — in one of two shapes.
*
* **A bin vector** is what a task produces: one value per bin over the whole
* assembly, on a grid the caller states. Every analysis output is this, because
* every analysis output is a measurement *per bin*.
*
* **Spans** are what a comparison of two loaded tracks is, and they are here
* because forcing that onto a grid was silently destroying data. A ratio of two
* base-resolution bigWigs is exact at every base — the lane draws it that way,
* cut at every boundary of either operand — and the export read both operands
* onto the contact map's bins first, so a reader who divided two 10 bp tracks
* on a 20 kb map got a 20 kb bigWig back and nothing said so. Nothing about the
* file format required it: a bedGraph section holds arbitrary intervals, which
* is decision 1 above, and the grid was an assumption in the *caller*.
*
* Everything past {@link intervalsFor} is identical for the two: sections, both
* trees, the summary and the zoom ladder all work on intervals and never learn
* which road produced them.
*/
export type BigWigInput = BigWigBins | BigWigIntervals
/** Values on a grid the caller states — every analysis output. */
export interface BigWigBins {
/** Sequences of the assembly, in the order the file lists them. */
contigs: readonly BigWigContig[]
/** One value per bin over the WHOLE assembly, NaN where there is none. */
values: Float32Array
/** Absolute bin index values[0] corresponds to. */
firstBin: number
binsize: number
spans?: undefined
}
/** Values on no grid at all — a comparison of two loaded tracks. */
export interface BigWigIntervals {
/** Sequences of the assembly, in the order the file lists them. */
contigs: readonly BigWigContig[]
/** The data, per sequence. A sequence with none may be left out. */
spans: readonly BigWigSpans[]
values?: undefined
}
const BIGWIG_MAGIC = 0x888ffc26
const BPT_MAGIC = 0x78ca8c91
const RTREE_MAGIC = 0x2468ace0
/** Version 4 is what every reader in circulation expects of a bigWig. */
const BBI_VERSION = 4
const HEADER_BYTES = 64
const ZOOM_HEADER_BYTES = 24
const TOTAL_SUMMARY_BYTES = 40
const BPT_HEADER_BYTES = 32
/** `chromId` u32 plus `chromSize` u32, which is what `valSize` announces. */
const BPT_VALUE_BYTES = 8
const RTREE_HEADER_BYTES = 48
const RTREE_LEAF_ITEM_BYTES = 32
const RTREE_INTERNAL_ITEM_BYTES = 24
const SECTION_HEADER_BYTES = 24
const BEDGRAPH_ITEM_BYTES = 12
const ZOOM_RECORD_BYTES = 32
/** Decision 1: bedGraph, so that a gap is a gap and a short final bin is short. */
const SECTION_TYPE_BEDGRAPH = 1
/**
* Items per node in both trees. 256 is what UCSC writes; the number matters
* only in that it is what makes the chromosome tree multi-level once an assembly
* passes it, which a metagenome does and a yeast genome does not.
*/
const TREE_FAN_OUT = 256
/** Items per section, and records per zoom block — UCSC's figure for both. */
const ITEMS_PER_SECTION = 1024
const ZOOM_RECORDS_PER_BLOCK = 1024
/** Decision 5, the whole ladder. */
const FIRST_REDUCTION_FACTOR = 10
const REDUCTION_STEP = 4
const MAX_ZOOM_LEVELS = 10
/**
* `reductionLevel` is a u32 in a zoom header, so a ladder that walked past this
* would wrap silently and advertise a level coarser than the genome as a level
* finer than a gene. The coarsest rung of a full ladder is 4^9 × 10 × the bin
* size, which passes 2^32 at any bin size over about 1,600 bases — well inside
* the range this app hands out.
*/
const MAX_REDUCTION = 0xffffffff
/**
* A rough compression ratio for {@link bigWigBytesEstimate} only, measured on
* the demo bigWigs. Never used to size anything the file depends on.
*/
const ASSUMED_COMPRESSION = 0.4
/** A contig, with the two things the writer needs that the caller did not give. */
interface PlannedContig {
name: string
length: number
/**
* The contig's id in the file, which is its rank in **name** order rather than
* its position in the caller's list. Every UCSC-written bigWig numbers contigs
* that way, the B+ tree has to be in collating order for a binary-searching
* reader to find anything in it, and the R-tree is sorted on the same key — so
* one order serves all three. The caller's order is preserved where it
* actually means something, which is the bin grid: {@link binOffset} below.
*/
id: number
/** Absolute index, in the caller's grid, of this contig's first bin. */
binOffset: number
binCount: number
}
/** One bin that had a value: exactly what goes into a bedGraph item. */
interface Interval {
start: number
end: number
value: number
}
/** Where a block sits and what it spans, which is all the R-tree indexes. */
interface IndexedBlock {
startChromId: number
startBase: number
endChromId: number
endBase: number
offset: number
size: number
}
/** The four bounds of an R-tree item, before it has a place in the file. */
interface Bounds {
startChromId: number
startBase: number
endChromId: number
endBase: number
}
/** One aggregated window at one reduction level. */
interface ZoomRecord {
chromId: number
start: number
end: number
validCount: number
min: number
max: number
sum: number
sumSquares: number
}
/** What the 40-byte total summary says about the whole file. */
interface Summary {
validCount: number
min: number
max: number
sum: number
sumSquares: number
}
// ---------------------------------------------------------------------------
// The assembly
// ---------------------------------------------------------------------------
/**
* Contigs in file order, each knowing where its bins are in the caller's vector.
*
* The two orders are separated here once and never again: the caller's order
* defines the bin grid, because `firstBin` counts bins along the contigs as the
* caller listed them, and name order defines the ids, because that is what the
* two indexes are sorted on. Conflating them would put every value on the wrong
* sequence in a file that still opens cleanly, which is the failure mode worth
* spending a paragraph to avoid.
*/
function planContigs(input: BigWigInput): PlannedContig[] {
const { contigs } = input
// A bin size only where there are bins. The grid checks below are about a
// *vector* — how long it is, and where its first element sits — and there is
// no vector on the spans road; the contig checks are about the assembly and
// apply to both.
const binsize = input.values === undefined ? 1 : input.binsize
if (!Number.isInteger(binsize) || binsize < 1) {
throw new Error(`A bigWig needs a whole bin size of at least one base; got ${binsize}.`)
}
if (contigs.length === 0) throw new Error('A bigWig needs at least one contig.')
const seen = new Set<string>()
const planned: PlannedContig[] = []
let binOffset = 0
for (const contig of contigs) {
if (contig.name === '') throw new Error('A contig with an empty name cannot be written.')
if (seen.has(contig.name)) throw new Error(`Two contigs are called ${contig.name}.`)
seen.add(contig.name)
if (!Number.isInteger(contig.length) || contig.length < 1) {
throw new Error(`${contig.name} has length ${contig.length}, which is not a sequence.`)
}
const binCount = input.values === undefined ? 0 : Math.ceil(contig.length / binsize)
planned.push({ name: contig.name, length: contig.length, id: 0, binOffset, binCount })
binOffset += binCount
}
if (input.values !== undefined) {
const { firstBin, values } = input
if (!Number.isInteger(firstBin) || firstBin < 0) {
throw new Error(`firstBin must be a whole, non-negative bin index; got ${firstBin}.`)
}
// A vector that runs past the last bin of the assembly means the caller's
// grid and this one disagree, and every value would then be attributed to
// the wrong place quietly. Cheaper to refuse than to explain later.
if (firstBin + values.length > binOffset) {
throw new Error(
`The value vector covers bins ${firstBin}–${firstBin + values.length} but the assembly ` +
`only has ${binOffset} bins at this bin size.`,
)
}
} else {
// The same class of disagreement on the other road: a span naming a sequence
// the assembly does not list has nowhere to go, and dropping it silently
// would write a file missing data the caller believes is in it.
const known = new Set(contigs.map((contig) => contig.name))
for (const span of input.spans) {
if (!known.has(span.contig)) {
throw new Error(`There are spans on ${span.contig}, which is not in the assembly.`)
}
if (span.starts.length !== span.ends.length || span.starts.length !== span.values.length) {
throw new Error(`${span.contig}'s starts, ends and values are of different lengths.`)
}
}
}
// Collated by UTF-8 bytes rather than by JavaScript's UTF-16 comparison,
// because the B+ tree stores keys as bytes and a reader binary-searches them
// as bytes. The two agree on ASCII contig names and part company above it.
const encoder = new TextEncoder()
const keys = new Map<string, Uint8Array>()
for (const contig of planned) keys.set(contig.name, encoder.encode(contig.name))
planned.sort((a, b) => compareBytes(keys.get(a.name)!, keys.get(b.name)!))
planned.forEach((contig, id) => {
contig.id = id
})
return planned
}
/** `memcmp`, with the shorter key first when one is a prefix of the other. */
function compareBytes(a: Uint8Array, b: Uint8Array): number {
const shared = Math.min(a.length, b.length)
for (let i = 0; i < shared; i += 1) {
if (a[i] !== b[i]) return a[i] - b[i]
}
return a.length - b.length
}
/**
* The bins of one contig that carry a value, as bedGraph intervals.
*
* Both halves of decision 1 are here and nowhere else. A `NaN` bin produces no
* interval at all, so a downstream tool asking about that range is told there is
* no data rather than handed a number that is not one; and the final interval's
* end is clamped to the sequence length, so a 7,500-base contig at a 1 kb bin
* size ends at 7,500 and not at 8,000. A fixedStep section could express neither
* without a second code path for each.
*
* **`±Infinity` is dropped here too, and not only `NaN`.** BBI has no way to
* write one: it would go into the total summary and into every zoom record over
* it, where `sumData` becomes `NaN` and the bounds become infinite — a file that
* opens cleanly in every reader with its autoscale and its mean destroyed, and
* which with only `+Infinity` present is not even visibly broken because `min`
* is still finite. That is strictly worse than the interval being absent, which
* is a state every reader already understands. A task that produces an infinity
* — a ratio with a zero denominator is the obvious way — has a bug of its own,
* and the honest place to fix it is the task; this is the backstop that keeps
* the bug out of the file.
*/
function intervalsFor(contig: PlannedContig, input: BigWigInput): Interval[] {
return input.values === undefined
? spansOf(contig, input.spans)
: binsOf(contig, input.values, input.firstBin, input.binsize)
}
function binsOf(
contig: PlannedContig,
values: Float32Array,
firstBin: number,
binsize: number,
): Interval[] {
const out: Interval[] = []
for (let bin = 0; bin < contig.binCount; bin += 1) {
const index = contig.binOffset + bin - firstBin
if (index < 0 || index >= values.length) continue
const value = values[index]
if (!Number.isFinite(value)) continue
const start = bin * binsize
out.push({ start, end: Math.min(start + binsize, contig.length), value })
}
return out
}
/**
* The same, for data that was never on a grid.
*
* Both halves of the rule above hold unchanged and for the same reasons: a value
* that is not finite produces no interval, and an end past the sequence is
* clamped to it. What is gone is the arithmetic — the caller's intervals are
* written as the caller's intervals, which is the whole point of this road.
*
* Sorted, and empty spans dropped. The R-tree is built over these in order and a
* bedGraph section header carries the first and last coordinate of its items, so
* an out-of-order interval is a file that opens and indexes wrongly. The caller
* that produced these has them in order already (a common refinement is a walk
* up the axis), which makes this a cheap check rather than a real sort.
*/
function spansOf(contig: PlannedContig, spans: readonly BigWigSpans[]): Interval[] {
const out: Interval[] = []
for (const span of spans) {
if (span.contig !== contig.name) continue
for (let at = 0; at < span.starts.length; at += 1) {
const value = span.values[at]
if (!Number.isFinite(value)) continue
const start = Math.max(0, span.starts[at])
const end = Math.min(contig.length, span.ends[at])
if (!(end > start)) continue
out.push({ start, end, value })
}
}
out.sort((one, two) => one.start - two.start)
return out
}
// ---------------------------------------------------------------------------
// Sections, summaries and zoom levels
// ---------------------------------------------------------------------------
/** The total summary, over every interval in the file. */
function summarise(perContig: Interval[][]): Summary {
let validCount = 0
let min = Infinity
let max = -Infinity
let sum = 0
let sumSquares = 0
for (const intervals of perContig) {
for (const item of intervals) {
// Weighted by the span, not counted per item: `validCount` in a BBI
// summary is bases covered, which the demo files confirm to the byte.
const span = item.end - item.start
validCount += span
sum += item.value * span
sumSquares += item.value * item.value * span
if (item.value < min) min = item.value
if (item.value > max) max = item.value
}
}
// An all-`NaN` vector is a legitimate input — a metacool contig nothing mapped
// to — and zeros are what a reader shows for it. `Infinity` in the header
// would be read as a real bound and would wreck an autoscale.
if (validCount === 0) return { validCount: 0, min: 0, max: 0, sum: 0, sumSquares: 0 }
return { validCount, min, max, sum, sumSquares }
}
/** One bedGraph section: header, then twelve bytes an item. */
function sectionBytes(chromId: number, items: Interval[]): Uint8Array {
const bytes = new Uint8Array(SECTION_HEADER_BYTES + items.length * BEDGRAPH_ITEM_BYTES)
const view = new DataView(bytes.buffer)
view.setUint32(0, chromId, true)
view.setUint32(4, items[0].start, true)
view.setUint32(8, items[items.length - 1].end, true)
// `itemStep` and `itemSpan` describe a regular grid and a bedGraph section has
// none — every item states its own bounds — so both are zero, which is what
// every reader takes them to mean for `type = 1`.
view.setUint32(12, 0, true)
view.setUint32(16, 0, true)
view.setUint8(20, SECTION_TYPE_BEDGRAPH)
view.setUint8(21, 0)
view.setUint16(22, items.length, true)
let p = SECTION_HEADER_BYTES
for (const item of items) {
view.setUint32(p, item.start, true)
view.setUint32(p + 4, item.end, true)
view.setFloat32(p + 8, item.value, true)
p += BEDGRAPH_ITEM_BYTES
}
return bytes
}
/**
* Every interval, cut into sections.
*
* A section never spans a contig, because its header carries a single
* `chromId`; beyond that the only rule is a cap on the item count, which is
* what keeps a reader's fetch for one window from dragging in a chromosome.
*/
function buildSections(
contigs: PlannedContig[],
perContig: Interval[][],
): { raw: Uint8Array[]; bounds: Bounds[] } {
const raw: Uint8Array[] = []
const bounds: Bounds[] = []
for (const contig of contigs) {
const intervals = perContig[contig.id]
for (let first = 0; first < intervals.length; first += ITEMS_PER_SECTION) {
const items = intervals.slice(first, first + ITEMS_PER_SECTION)
raw.push(sectionBytes(contig.id, items))
bounds.push({
startChromId: contig.id,
startBase: items[0].start,
endChromId: contig.id,
endBase: items[items.length - 1].end,
})
}
}
return { raw, bounds }
}
/**
* The whole zoom ladder, records and all, in ascending order of reduction.
*
* Decision 5. Ten times the bin size for the first level, because anything finer
* is a level a reader would never choose over the real data, then four times per
* level so that consecutive levels are far enough apart to be worth storing
* separately, and ten levels at the outside.
*
* Two things stop it early, and the second was put here by a metagenome. The
* first is the stated rule: a level holding fewer records than a tree node holds
* items is the last one worth having, since below that the index costs more than
* the records it points at and a reader drawing the whole genome reads one block
* either way. The second is that **a level which reduces nothing is not a
* level**. A zoom window never spans two contigs, so on an assembly of twenty
* thousand short contigs every reduction past the contig length yields exactly
* one record per contig for ever — the first rule never fires, and the file ends
* up with nine identical copies of the same 20,000 records and a `reductionLevel`
* that has wrapped past 2^32. Refusing a level that is no smaller than the one
* below it deletes both problems with one comparison.
*/
/**
* The span the ladder's first rung is ten times, which on a grid is the bin.
*
* Off the grid there is no one span, so it is the **mean** of what was actually
* written — the same figure `bedGraphToBigWig` reduces from, and the right one:
* the ladder exists so that a reader zoomed out fetches a summary instead of a
* million items, and how far out that starts mattering is decided by how big the
* items are on average rather than by the smallest or the largest of them.
*
* Never below one. A file of single-base intervals has a mean span of exactly
* one, and a reduction of zero would make the first rung the same resolution as
* the data and every rung after it identical.
*/
function baseSpanOf(input: BigWigInput, perContig: Interval[][]): number {
if (input.values !== undefined) return input.binsize
let bases = 0
let items = 0
for (const intervals of perContig) {
for (const item of intervals) {
bases += item.end - item.start
items += 1
}
}
return items === 0 ? 1 : Math.max(1, Math.round(bases / items))
}
function zoomLadder(
contigs: PlannedContig[],
perContig: Interval[][],
binsize: number,
): { reduction: number; records: ZoomRecord[] }[] {
const out: { reduction: number; records: ZoomRecord[] }[] = []
let previous = Infinity
let reduction = binsize * FIRST_REDUCTION_FACTOR
for (let level = 0; level < MAX_ZOOM_LEVELS && reduction <= MAX_REDUCTION; level += 1) {
const records = zoomRecords(contigs, perContig, reduction)
if (records.length === 0 || records.length >= previous) break
out.push({ reduction, records })
previous = records.length
if (records.length < TREE_FAN_OUT) break
reduction *= REDUCTION_STEP
}
return out
}
/**
* One zoom level, aggregated straight from the base intervals.
*
* Never from the level below: an aggregation of an aggregation carries the
* rounding of every level it passed through, and since `sumData` and
* `sumSquares` are stored as float32 in a zoom record that error is real and
* compounds. Recomputing from the base vector costs one pass over a few thousand
* intervals per level, which is nothing.
*
* Every reduction is a whole multiple of the bin size by construction, so a bin
* lies entirely inside one window and the windows are visited in order — which
* is why one open record is enough and no sorting happens here.
*/
function zoomRecords(
contigs: PlannedContig[],
perContig: Interval[][],
reduction: number,
): ZoomRecord[] {
const out: ZoomRecord[] = []
for (const contig of contigs) {
let open: ZoomRecord | null = null
for (const item of perContig[contig.id]) {
const window = Math.floor(item.start / reduction)
const start = window * reduction
if (open === null || open.start !== start) {
if (open !== null) out.push(open)
open = {
chromId: contig.id,
start,
// Clamped for the same reason the last bin of a contig is: a summary
// that claims bases the sequence does not have is a summary a browser
// draws past the end of the chromosome.
end: Math.min(start + reduction, contig.length),
validCount: 0,
min: Infinity,
max: -Infinity,
sum: 0,
sumSquares: 0,
}
}
const span = item.end - item.start
open.validCount += span
open.sum += item.value * span
open.sumSquares += item.value * item.value * span
if (item.value < open.min) open.min = item.value
if (item.value > open.max) open.max = item.value
}
if (open !== null) out.push(open)
}
return out
}
/** Zoom records packed into blocks of at most {@link ZOOM_RECORDS_PER_BLOCK}. */
function buildZoomBlocks(records: ZoomRecord[]): { raw: Uint8Array[]; bounds: Bounds[] } {
const raw: Uint8Array[] = []
const bounds: Bounds[] = []
for (let first = 0; first < records.length; first += ZOOM_RECORDS_PER_BLOCK) {
const slice = records.slice(first, first + ZOOM_RECORDS_PER_BLOCK)
const bytes = new Uint8Array(slice.length * ZOOM_RECORD_BYTES)
const view = new DataView(bytes.buffer)
let p = 0
for (const record of slice) {
view.setUint32(p, record.chromId, true)
view.setUint32(p + 4, record.start, true)
view.setUint32(p + 8, record.end, true)
view.setUint32(p + 12, record.validCount, true)
view.setFloat32(p + 16, record.min, true)
view.setFloat32(p + 20, record.max, true)
view.setFloat32(p + 24, record.sum, true)
view.setFloat32(p + 28, record.sumSquares, true)
p += ZOOM_RECORD_BYTES
}
raw.push(bytes)
// Unlike a section, a zoom block may straddle contigs — each record carries
// its own `chromId`, so nothing in the format forbids it, and forbidding it
// anyway would cost a metagenome one block per contig per level.
bounds.push({
startChromId: slice[0].chromId,
startBase: slice[0].start,
endChromId: slice[slice.length - 1].chromId,
endBase: slice[slice.length - 1].end,
})
}
return { raw, bounds }
}
// ---------------------------------------------------------------------------
// The two indexes
// ---------------------------------------------------------------------------
/** The chromosome B+ tree: name to id and size, collated, multi-level. */
function chromosomeTreeBytes(contigs: PlannedContig[], treeOffset: number): Uint8Array {
const encoder = new TextEncoder()
const keys = contigs.map((contig) => encoder.encode(contig.name))
// One key width for the whole tree, so a node is a fixed stride. Zero-padded,
// and a reader stops at the first NUL.
const keySize = keys.reduce((widest, key) => Math.max(widest, key.length), 1)
const itemBytes = keySize + BPT_VALUE_BYTES
const plan = planTree(contigs.length, {
fanOut: TREE_FAN_OUT,
leafItemBytes: itemBytes,
// An internal item is a key and an eight-byte child offset, which happens to
// be the same width as a leaf item here; they are still named separately
// because nothing in the format requires them to agree.
internalItemBytes: keySize + 8,
rootOffset: treeOffset + BPT_HEADER_BYTES,
})
const bytes = new Uint8Array(BPT_HEADER_BYTES + plan.bytes)
const view = new DataView(bytes.buffer)
view.setUint32(0, BPT_MAGIC, true)
view.setUint32(4, TREE_FAN_OUT, true)
view.setUint32(8, keySize, true)
view.setUint32(12, BPT_VALUE_BYTES, true)
view.setBigUint64(16, BigInt(contigs.length), true)
view.setBigUint64(24, 0n, true)
for (const node of plan.nodes) {
let p = node.offset - treeOffset
view.setUint8(p, node.isLeaf ? 1 : 0)
view.setUint8(p + 1, 0)
view.setUint16(p + 2, node.isLeaf ? node.leafCount : node.children.length, true)
p += TREE_NODE_HEADER_BYTES
if (node.isLeaf) {
for (let i = 0; i < node.leafCount; i += 1) {
const contig = contigs[node.leafFirst + i]
bytes.set(keys[node.leafFirst + i], p)
view.setUint32(p + keySize, contig.id, true)
view.setUint32(p + keySize + 4, contig.length, true)
p += itemBytes
}
} else {
for (const childIndex of node.children) {
const child = plan.nodes[childIndex]
// The key an internal item carries is the first key in the subtree it
// points at, which is what makes the descent a comparison rather than a
// scan.
bytes.set(keys[child.leafFirst], p)
view.setBigUint64(p + keySize, BigInt(child.offset), true)
p += keySize + 8
}
}
}
return bytes
}
/** The bounding box of a run of blocks, folded rather than argued about. */
function boundsOver(blocks: IndexedBlock[], first: number, count: number): Bounds {
const box: Bounds = {
startChromId: blocks[first].startChromId,
startBase: blocks[first].startBase,
endChromId: blocks[first].endChromId,
endBase: blocks[first].endBase,
}
for (let i = 1; i < count; i += 1) {
const block = blocks[first + i]
if (
block.startChromId < box.startChromId ||
(block.startChromId === box.startChromId && block.startBase < box.startBase)
) {
box.startChromId = block.startChromId
box.startBase = block.startBase
}
if (
block.endChromId > box.endChromId ||
(block.endChromId === box.endChromId && block.endBase > box.endBase)
) {
box.endChromId = block.endChromId
box.endBase = block.endBase
}
}
return box
}
/**
* The R-tree over a list of blocks, which must already be in `(chromId, start)`
* order — they are, because the sections are generated contig by contig in id
* order and the zoom records follow the same walk.
*
* `endFileOffset` is written as the offset one past the last indexed block,
* which is where this index itself begins. The field is advisory and nothing in
* this repo's reader consults it — `Yeast_ATAC.bw` carries a value there that
* cannot possibly be right, which is a reason to write a truthful one and no
* reason to depend on anyone else's.
*/
function rTreeBytes(blocks: IndexedBlock[], treeOffset: number, itemsPerSlot: number): Uint8Array {
const plan = planTree(blocks.length, {
fanOut: TREE_FAN_OUT,
leafItemBytes: RTREE_LEAF_ITEM_BYTES,
internalItemBytes: RTREE_INTERNAL_ITEM_BYTES,
rootOffset: treeOffset + RTREE_HEADER_BYTES,
})
const bytes = new Uint8Array(RTREE_HEADER_BYTES + plan.bytes)
const view = new DataView(bytes.buffer)
const whole =
blocks.length === 0
? { startChromId: 0, startBase: 0, endChromId: 0, endBase: 0 }
: boundsOver(blocks, 0, blocks.length)
view.setUint32(0, RTREE_MAGIC, true)
view.setUint32(4, TREE_FAN_OUT, true)
view.setBigUint64(8, BigInt(blocks.length), true)
view.setUint32(16, whole.startChromId, true)
view.setUint32(20, whole.startBase, true)
view.setUint32(24, whole.endChromId, true)
view.setUint32(28, whole.endBase, true)
view.setBigUint64(32, BigInt(treeOffset), true)
view.setUint32(40, itemsPerSlot, true)
// The reserved tail of an R-tree header is a u32, not a u64: the header is 48
// bytes and the root node starts immediately after it, which is where
// `@gmod/bbi` looks for it.
view.setUint32(44, 0, true)
for (const node of plan.nodes) {
let p = node.offset - treeOffset
view.setUint8(p, node.isLeaf ? 1 : 0)
view.setUint8(p + 1, 0)
view.setUint16(p + 2, node.isLeaf ? node.leafCount : node.children.length, true)
p += TREE_NODE_HEADER_BYTES
if (node.isLeaf) {
for (let i = 0; i < node.leafCount; i += 1) {
const block = blocks[node.leafFirst + i]
view.setUint32(p, block.startChromId, true)
view.setUint32(p + 4, block.startBase, true)
view.setUint32(p + 8, block.endChromId, true)
view.setUint32(p + 12, block.endBase, true)
view.setBigUint64(p + 16, BigInt(block.offset), true)
view.setBigUint64(p + 24, BigInt(block.size), true)
p += RTREE_LEAF_ITEM_BYTES
}
} else {
for (const childIndex of node.children) {
const child = plan.nodes[childIndex]
const box = boundsOver(blocks, child.leafFirst, child.leafCount)
view.setUint32(p, box.startChromId, true)
view.setUint32(p + 4, box.startBase, true)
view.setUint32(p + 8, box.endChromId, true)
view.setUint32(p + 12, box.endBase, true)
view.setBigUint64(p + 16, BigInt(child.offset), true)
p += RTREE_INTERNAL_ITEM_BYTES
}
}
}
return bytes
}
// ---------------------------------------------------------------------------
// Compression
// ---------------------------------------------------------------------------
/**
* Every block of the file, deflated as a unit — decision 2.
*
* `deflateBytes` is `ui/export.ts`'s, unchanged: it wraps the platform's
* `CompressionStream('deflate')`, whose output is RFC 1950 zlib framing, which
* is exactly what a BBI section is. It reports a `null` filter where the
* platform has no `CompressionStream`, and that verdict has to be the same for
* every block in the file, because `uncompressBufSize` is one file-wide switch
* rather than a per-block flag — hence one pass over everything here and one
* answer out.
*/
async function packBlocks(
groups: Uint8Array[][],
deflate?: (bytes: Uint8Array) => Uint8Array,
): Promise<{ packed: Uint8Array[][]; compressed: boolean; largest: number }> {
const packed: Uint8Array[][] = []
let compressed = true
let largest = 0
for (const group of groups) {
const out: Uint8Array[] = []
for (const bytes of group) {
largest = Math.max(largest, bytes.length)
if (deflate !== undefined) {
out.push(deflate(bytes))
continue
}
const { data, filter } = await deflateBytes(bytes)
if (filter === null) compressed = false
out.push(data)
}
packed.push(out)
}
return { packed, compressed, largest }
}
/** How {@link writeBigWig} may be told to compress. */
export interface BigWigWriting {
/**
* zlib's RFC 1950 deflate, synchronous, where the caller has one — the command
* line has Node's. The platform's `CompressionStream` is the same zlib at the
* same settings, so the bytes are the same; what it costs is a stream, a blob
* and a response per block, and a file is hundreds of blocks. Absent, every
* block goes through `CompressionStream` (decision 2).
*/
deflate?: (bytes: Uint8Array) => Uint8Array
}
// ---------------------------------------------------------------------------
// The file
// ---------------------------------------------------------------------------
/**
* The whole bigWig, as a blob ready to be saved.
*
* Assembled by decision 4's accumulator: chunks are pushed into a list with a
* running length, so the length *is* the offset of whatever goes next, and the
* header and zoom headers are pushed first as blank slots and filled in at the
* end against offsets that are known by then. A genome's worth of bins is
* kilobytes, so there is nothing to stream and nothing to be clever about.
*/
export async function writeBigWig(input: BigWigInput, writing: BigWigWriting = {}): Promise<Blob> {
const contigs = planContigs(input)
const perContig: Interval[][] = new Array<Interval[]>(contigs.length)
for (const contig of contigs) perContig[contig.id] = intervalsFor(contig, input)
const summary = summarise(perContig)
const sections = buildSections(contigs, perContig)
const zooms = zoomLadder(contigs, perContig, baseSpanOf(input, perContig)).map((level) => ({
reduction: level.reduction,
...buildZoomBlocks(level.records),
}))
const { packed, compressed, largest } = await packBlocks(
[sections.raw, ...zooms.map((zoom) => zoom.raw)],
writing.deflate,
)
const sectionBlocks = packed[0]
const zoomBlocks = packed.slice(1)
const chunks: Uint8Array[] = []
let length = 0
const push = (bytes: Uint8Array): void => {
chunks.push(bytes)
length += bytes.length
}
const header = new Uint8Array(HEADER_BYTES)
const headerView = new DataView(header.buffer)
push(header)
const zoomHeaders = new Uint8Array(ZOOM_HEADER_BYTES * zooms.length)
const zoomHeaderView = new DataView(zoomHeaders.buffer)
push(zoomHeaders)
const totalSummaryOffset = length
push(summaryBytes(summary))
const chromosomeTreeOffset = length
push(chromosomeTreeBytes(contigs, chromosomeTreeOffset))
// The u64 here is the **section** count, verified against the demo bigWigs;
// the blocks themselves start eight bytes later.
const fullDataOffset = length
push(u64(sectionBlocks.length))
const sectionIndex = placeBlocks(sectionBlocks, sections.bounds, push, () => length)
const fullIndexOffset = length
// One section per leaf slot, which is what `itemsPerSlot = 1` means in an
// unzoomed index — not the 1024 items a section holds.
push(rTreeBytes(sectionIndex, fullIndexOffset, 1))
const placed = zooms.map((zoom, level) => {
// A zoom level's count is a u32 and holds the level's **block** count, not
// its record count — the asymmetry with the u64 above is what the demo files
// actually carry.
const dataOffset = length
push(u32(zoomBlocks[level].length))
const index = placeBlocks(zoomBlocks[level], zoom.bounds, push, () => length)
const indexOffset = length
push(rTreeBytes(index, indexOffset, ZOOM_RECORDS_PER_BLOCK))
return { reduction: zoom.reduction, dataOffset, indexOffset }
})
// The magic again, so that a reader can tell a complete file from a download
// that stopped early.
push(u32(BIGWIG_MAGIC))
headerView.setUint32(0, BIGWIG_MAGIC, true)
headerView.setUint16(4, BBI_VERSION, true)
headerView.setUint16(6, zooms.length, true)
headerView.setBigUint64(8, BigInt(chromosomeTreeOffset), true)
headerView.setBigUint64(16, BigInt(fullDataOffset), true)
headerView.setBigUint64(24, BigInt(fullIndexOffset), true)
// A bigWig has no BED columns and no autoSql schema; all three fields are the
// zero that says so.
headerView.setUint16(32, 0, true)
headerView.setUint16(34, 0, true)
headerView.setBigUint64(36, 0n, true)
headerView.setBigUint64(44, BigInt(totalSummaryOffset), true)
// The largest **uncompressed** block, which is the buffer a reader allocates
// before inflating one. Zero is the signal that nothing was compressed.
headerView.setUint32(52, compressed ? largest : 0, true)
headerView.setBigUint64(56, 0n, true)
placed.forEach((zoom, level) => {
const p = level * ZOOM_HEADER_BYTES
zoomHeaderView.setUint32(p, zoom.reduction, true)
zoomHeaderView.setUint32(p + 4, 0, true)
zoomHeaderView.setBigUint64(p + 8, BigInt(zoom.dataOffset), true)
zoomHeaderView.setBigUint64(p + 16, BigInt(zoom.indexOffset), true)
})
return new Blob(chunks as BlobPart[], { type: 'application/octet-stream' })
}
/** Push a run of blocks, recording where each one landed for the index. */
function placeBlocks(
blocks: Uint8Array[],
bounds: Bounds[],
push: (bytes: Uint8Array) => void,
here: () => number,
): IndexedBlock[] {
const indexed: IndexedBlock[] = []
blocks.forEach((block, i) => {
const offset = here()
push(block)
indexed.push({ ...bounds[i], offset, size: block.length })
})
return indexed
}
/** The 40-byte total summary. */
function summaryBytes(summary: Summary): Uint8Array {
const bytes = new Uint8Array(TOTAL_SUMMARY_BYTES)
const view = new DataView(bytes.buffer)
view.setBigUint64(0, BigInt(summary.validCount), true)
view.setFloat64(8, summary.min, true)
view.setFloat64(16, summary.max, true)
view.setFloat64(24, summary.sum, true)
view.setFloat64(32, summary.sumSquares, true)
return bytes
}
function u32(value: number): Uint8Array {
const bytes = new Uint8Array(4)
new DataView(bytes.buffer).setUint32(0, value, true)
return bytes
}
function u64(value: number): Uint8Array {
const bytes = new Uint8Array(8)
new DataView(bytes.buffer).setBigUint64(0, BigInt(value), true)
return bytes
}
// ---------------------------------------------------------------------------
// The estimate
// ---------------------------------------------------------------------------
/**
* Roughly how many bytes {@link writeBigWig} would produce, for a UI hint.
*
* Rough on purpose, and stated as such wherever it is shown: the parts that can
* be counted exactly are counted, and the one thing that cannot — how well the
* values deflate — is a single measured constant. A track of smooth coverage
* beats it and a track of noise misses it, in both cases by less than the order
* of magnitude the hint exists to convey. Cheap enough to call while a slider
* moves: one pass over the vector and no allocation per bin.
*/
export function bigWigBytesEstimate(input: BigWigInput): number {
const contigs = planContigs(input)
// Counted the way the writer counts, on whichever road the caller took, so an
// estimate and a file can never disagree about how many items there are.
let items = 0
if (input.values === undefined) {
for (const span of input.spans) {
for (let at = 0; at < span.values.length; at += 1) {
if (Number.isFinite(span.values[at])) items += 1
}
}
} else {
for (const contig of contigs) {
for (let bin = 0; bin < contig.binCount; bin += 1) {
const index = contig.binOffset + bin - input.firstBin
if (index < 0 || index >= input.values.length) continue
if (!Number.isNaN(input.values[index])) items += 1
}
}
}
if (items === 0) return HEADER_BYTES + TOTAL_SUMMARY_BYTES + BPT_HEADER_BYTES + RTREE_HEADER_BYTES
const sections = Math.ceil(items / ITEMS_PER_SECTION)
const data = sections * SECTION_HEADER_BYTES + items * BEDGRAPH_ITEM_BYTES
// The ladder reduces by ten and then by four, so its records are a geometric
// series summing to about a third of a tenth of the base items.
const zoomRecordCount = Math.ceil(items / FIRST_REDUCTION_FACTOR) * (REDUCTION_STEP / 3)
const zoomData = zoomRecordCount * ZOOM_RECORD_BYTES
const indexes =
(sections + Math.ceil(zoomRecordCount / ZOOM_RECORDS_PER_BLOCK)) * RTREE_LEAF_ITEM_BYTES * 1.1
const chromTree = contigs.length * 40
return Math.round(
HEADER_BYTES +
MAX_ZOOM_LEVELS * ZOOM_HEADER_BYTES +
TOTAL_SUMMARY_BYTES +
chromTree +
indexes +
(data + zoomData) * ASSUMED_COMPRESSION,
)
}
|