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 | 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 13x 13x 13x 13x 13x 13x 13x 13x 13x 6x 6x 6x 2x 4x 4x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 13x 2x 2x 2x 2x 2x 2x 2x 10x 10x 10x 10x 10x 10x 10x 13x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 13x 5x 5x 5x 5x 5x 5x 10x 10x 13x 13x 13x 13x 7x 1x 1x 1x 1x 7x 7x 6x 13x 13x 5x 13x 5x 5x 5x 5x 5x 5x 2x 2x 2x 2x 32248x 32248x 32248x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 32248x 32248x 6x 32248x 6x 6x 1x 5x 16124x 16124x 16124x 16124x 16124x 16124x 16124x 16124x 16124x 16124x 16124x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 6x 2x 2x 2x 2x 6x 6x 6x 6x 6x 6x 5x 5x 5x 5x 5x 5x 5x 5x 5x 7x 7x 7x 5x 5x 5x 5x 5x 5x 5x 16x 16x 16x 16x 16x 4x 4x 4x 4x 4x 4x 7x 7x 7x 7x 7x 7x 7x 7x 2x 5x 1x 4x 3x 1x 7x 1x 1x 1x 1x 1x 1x 6x 6x 7x 7x 2x 23x 23x 23x 23x 23x 23x 23x 23x 23x 13x 13x 13x 23x 23x 23x 23x 23x 23x 23x 23x 23x 23x 23x 23x 23x 2x 10x 10x 10x 10x 10x 10x 10x 10x 10x 30x 30x 30x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 10x 5x 5x 5x 5x 5x 5x 2x 3x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 20x 20x 20x 20x 20x 20x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 20x 20x 20x 5x 5x 5x 30x 30x 30x 30x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 355x 117x 117x 117x 355x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 2x 2x 5x 5x 5x 5x 5x 5x 5x 5x 7x 7x 7x 5x 5x 5x 5x 7x 7x 5x 5x 7x 878x 878x 878x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 300x 300x 240x 240x 5x 5x 5x 5x 5x 5x 42x 42x 42x 42x 42x 42x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x | /**
* Pairs QC, run: the worker-side half of `analysis/tasks/pairsqc.ts`.
*
* Three readings of one library's records, each from wherever it is cheapest to
* have them:
*
* - **`pairtools stats`** — records, mapping, duplicates, cis and trans, pair
* types — drawn as figures and written as the table `pairtools stats` writes
* (`pairs/stats.ts` is the port, held to pairtools by its golden);
* - **P(s) by orientation**, the four curves whose convergence is the shortest
* separation the library can be believed at (`analysis/orientation.ts`);
* - **restriction fragments**, with a reference and an enzyme: which pairs are a
* fragment ligated to itself or to its neighbour rather than a contact, how
* far each read ended from a cut, how many fragments the library saw, and how
* coverage follows the sites (`genome/restriction.ts` is the arithmetic).
*
* ## One pass, at most
*
* A map binned from a `.pairs` had the first two counted by its import (§3.7)
* and they arrive on `TaskInputs` with nothing read. The third needs the
* records again — the import could not have placed a read in a fragment it did
* not know about — so it is one pass over the map's own file
* (`TaskInputs.mapSource`). A map opened as a matrix kept no records, and all
* three come from one pass over the `.pairs` the form names.
*
* Either pass is **the import's own record reader** (`pairs/bin.ts`'s
* `binPairs`), run at one bin per sequence so its binning costs nothing, with
* the restriction counter riding it through `onRecord`: one scan feeds the
* stats, the orientation curves and the fragments, and §3.7's column handling
* is not written a second time.
*/
import { contigIndexOf } from '../../genome/naming'
import {
ORIENTATIONS,
PAIR_CLASSES,
RestrictionCounter,
cutsOfResult,
enzymeLabel,
enzymesFor,
type CutSites,
type RestrictionStats,
} from '../../genome/restriction'
import { binPairs, type RecordObserver } from '../../pairs/bin'
import { readPairsHeader, type PairsContig } from '../../pairs/format'
import {
convergenceOf,
pairtoolsStatsEntries,
statText,
type PairStats,
} from '../../pairs/stats'
import type { ByteSource } from '../../types'
import type { SourceSpec } from '../../worker/protocol'
import { binContigMap, CoverageAccumulator } from '../coverage'
import { numParam, strParam, type JobContext, type TaskRunner } from '../job'
import { orientationFigure, type OrientationCounts } from '../orientation'
import { spearman } from '../stats'
import { namedPairs, restrictionParams } from '../tasks/pairsqc'
import type { CurveOutput, CurveSeries, TableOutput, TaskOutput, TaskResult } from '../types'
// ---------------------------------------------------------------------------
// The runner
// ---------------------------------------------------------------------------
export const runPairsQc: TaskRunner = async (job) => {
const named = namedPairs(job.params)
const own = job.inputs.mapSource
const cuts = digestOf(job)
// What the import counted, where the records are the map's own: a `.pairs`
// named on the form is another file, and its counts are its own.
let stats: PairStats | null = named === '' ? (job.inputs.pairsStats ?? null) : null
let orientation: OrientationCounts | null = named === '' ? (job.inputs.orientation ?? null) : null
let restriction: RestrictionStats | null = null
let from = `as ${fileName(job.framing.uri)} was binned`
// A pass, for the two reasons there are: a file the import never read, and a
// digest the import knew nothing of.
const pass = named !== '' || cuts !== null || (stats === null && own !== undefined)
if (pass) {
const where = named !== '' ? named : fileName(job.framing.uri)
const source =
named !== ''
? await openRecords(job, named, job.inputs.sources?.['pairs'])
: own !== undefined
? await openOwn(job, own, where)
: null
if (source !== null) {
try {
const read = await readRecords(job, where, source, cuts)
if (read !== null) {
// The import's counts where it made them, which are the same records
// counted the same way; the pass's where it did not.
if (stats === null && read.stats !== null) from = `in one pass over ${where}`
stats ??= read.stats
orientation ??= read.orientation
restriction = read.restriction
}
} finally {
// The caller's to close (`JobContext.openSource`): a contact list left
// mounted after its one pass is memory the map could have had.
source.close()
}
}
}
if (stats === null && orientation === null && restriction === null) {
if (named === '' && own === undefined && !pass) {
job.note(
'This map was opened as a matrix, which keeps no records, so there is nothing to count. ' +
'Name the .pairs it was binned from.',
)
}
return []
}
const outputs: TaskOutput[] = []
if (orientation !== null) {
// The convergence said is pairtools', from the same records' `distFreq` —
// the number `pairsqc-stats` prints and the Pairs statistics card shows —
// and never the figure's own, which is a second rule with a second answer.
const figure = orientationOutput(job, orientation, 'pairsqc-orientation', { converge: false })
if (figure !== null) {
outputs.push(figure)
if (stats !== null) job.note(convergenceNote(stats))
}
} else if (stats !== null) {
job.note(
'The records carry no strand1/strand2 columns, so there is no P(s) by orientation: which ' +
'way a read pointed is the whole of what that figure is about.',
)
}
if (stats !== null) {
if (stats.scoped) {
job.note('The import read part of the file, so these counts are of the part it read.')
}
const types = typesOutput(stats)
if (types !== null) outputs.push(types)
else job.note('The records carry no pair_type column, so pair types are not counted.')
outputs.push(cisOutput(stats), statsOutput(stats))
job.note(
`The pairtools statistics were counted ${from}: ${stats.total.toLocaleString('en-US')} ` +
`records, ${stats.totalNodups.toLocaleString('en-US')} of them mapped on both sides and ` +
'not duplicates.',
)
}
if (restriction !== null) {
outputs.push(...restrictionOutputs(job, restriction))
if (cuts !== null) {
const sites = await coverageAgainstSites(job, cuts, restriction.enzymes)
if (sites !== null) outputs.push(sites)
}
}
return outputs
}
// ---------------------------------------------------------------------------
// The digest
// ---------------------------------------------------------------------------
/**
* The cuts the restriction step found, or `null` where there was no step.
*
* `null` with a sentence wherever the reader asked for one and it is not here,
* because a run that quietly left the restriction half off a figure the form
* asked for is a run that reads as a library with nothing to say about it.
*/
function digestOf(job: JobContext): Map<string, CutSites> | null {
const asked = restrictionParams(job.params)
const result: TaskResult | undefined = job.inputs.upstream?.['restriction']
if (asked === null) {
if (strParam(job.params, 'enzyme', 'none') !== 'none') {
job.note(
'An enzyme was chosen with no reference to find its sites in, so there are no ' +
'restriction statistics.',
)
}
return null
}
// A digest the enzymes cannot be read for is a refusal and not a note: the
// form never offers the Run (`tasks/pairsqc.ts`'s estimate says why), and a
// stored recipe or a command line that reaches here asked for something that
// cannot be done — answering it with the pairtools half alone would read as a
// library with nothing to say about its fragments.
const chosen = enzymesFor(strParam(asked, 'enzyme', 'none'), strParam(asked, 'sites', ''))
if (!Array.isArray(chosen)) throw new Error(chosen.refusal)
if (result === undefined) {
job.note('The restriction step did not run, so there are no restriction statistics.')
return null
}
const cuts = cutsOfResult(result)
if (cuts === null) {
job.note('The restriction step found no sequence to cut, so there are no restriction statistics.')
return null
}
return cuts
}
/** The enzymes a run named, as one word — `DpnII`, `DpnII+HinfI` — for the labels. */
function enzymesNamed(job: JobContext): string {
const chosen = enzymesFor(strParam(job.params, 'enzyme', 'none'), strParam(job.params, 'sites', ''))
return Array.isArray(chosen) ? enzymeLabel(chosen) : ''
}
// ---------------------------------------------------------------------------
// The records
// ---------------------------------------------------------------------------
const PLUS = 0x2b
const MINUS = 0x2d
const BANG = 0x21
const D = 0x44
/** A strand byte as the counter reads it, without a string made per record. */
function strandWord(byte: number): string {
return byte === PLUS ? '+' : byte === MINUS ? '-' : '.'
}
/**
* One pass over a `.pairs`: what `pairtools stats` counts, the four orientation
* curves, and — with a digest — the restriction fragments, all from one scan.
*
* **Through `binPairs` at the coarsest possible grid**, as P(s)'s own read of a
* named file is (`run/expected.ts`): one bin per sequence makes the binning
* free, and what the pass is for is the counters it carries. The restriction
* counter rides it through `onRecord`, which hands each record over exactly as
* the stats counter is handed it — the header's own name strings for every
* sequence it declares, so a record allocates nothing, and a name decoded only
* where the header lacks it.
*/
async function readRecords(
job: JobContext,
where: string,
source: ByteSource,
cuts: Map<string, CutSites> | null,
): Promise<{
stats: PairStats | null
orientation: OrientationCounts | null
restriction: RestrictionStats | null
} | null> {
job.progress(0, 1, 'reading records')
const header = await readPairsHeader(source)
const counter = cuts === null ? null : counterFor(job, cuts, header.contigs)
const names = header.contigs.map((contig) => contig.name)
const decoder = new TextDecoder()
const nameOf = (line: Uint8Array, c: number, from: number, to: number): string => {
if (c >= 0) return names[c]
if (to - from === 1 && line[from] === BANG) return '!'
return decoder.decode(line.subarray(from, to))
}
const onRecord: RecordObserver | undefined =
counter === null
? undefined
: (line, c1, c1From, c1To, c2, c2From, c2To, p1, p2, s1, s2, tFrom, tTo) => {
const duplicate = tTo - tFrom === 2 && line[tFrom] === D && line[tFrom + 1] === D
counter.add(
nameOf(line, c1, c1From, c1To),
p1 < 0 ? Number.NaN : p1,
strandWord(s1),
nameOf(line, c2, c2From, c2To),
p2 < 0 ? Number.NaN : p2,
strandWord(s2),
duplicate,
)
}
let longest = 0
for (const contig of header.contigs) longest = Math.max(longest, contig.length)
const binned = await binPairs(source, {
// One bin per sequence: the cheapest grid that exists and is not a map.
binSize: Math.max(1, longest),
contigs: header.contigs,
columns: header.columns,
onProgress: (done, total) => job.progress(done, total, 'reading records'),
cancelled: () => job.cancelled(),
...(onRecord === undefined ? {} : { onRecord }),
})
job.throwIfCancelled()
if (binned === null) return null
if (binned.skippedUnknownContig > 0 && counter !== null) {
// Said because it is the one place the two counts part: pairtools stats
// counts these with the rest, and the fragments cannot place them.
job.note(
`${binned.skippedUnknownContig.toLocaleString('en-US')} records of ${where} name a sequence ` +
'its header does not declare.',
)
}
return {
stats: binned.stats ?? null,
orientation: binned.orientation ?? null,
restriction: counter === null ? null : counter.result(),
}
}
/**
* The counter for this digest, with the lengths it cuts each sequence to.
*
* The cuts are the reference's, by its names; the lengths are the `.pairs`
* header's and then the map's, since the analysis worker holds no reference to
* ask — and a library's header declares the very assembly its reads were mapped
* to. Names are matched across spellings by the counter (§5.9). A reference
* sequence neither declares is left out of the fragments, and said to be.
*/
function counterFor(
job: JobContext,
cuts: Map<string, CutSites>,
declared: readonly PairsContig[],
): RestrictionCounter {
const lengths = new Map<string, number>()
for (const contig of declared) lengths.set(contig.name, contig.length)
const mc = job.mc
for (const name of mc.chromnames) {
const length = mc.chromsizes.get(name)
if (length !== undefined && !lengths.has(name)) lengths.set(name, length)
}
const known = [...lengths.keys()]
const unplaced = [...cuts.keys()].filter(
(name) => !lengths.has(name) && contigIndexOf(known, name) < 0,
)
if (unplaced.length > 0) {
job.note(
`${unplaced.length.toLocaleString('en-US')} of the reference's sequences ` +
`(${unplaced.slice(0, 3).join(', ')}${unplaced.length > 3 ? ', …' : ''}) are in neither ` +
'the .pairs header nor the map, so their fragments are not counted.',
)
}
return new RestrictionCounter(cuts, lengths, enzymesNamed(job))
}
/** A path's or an address's last part, for a sentence. */
function fileName(uri: string): string {
const path = uri.split('::')[0].split(/[?#]/)[0]
const last = path.split(/[\\/]/).filter((part) => part !== '').pop()
return last ?? uri
}
/**
* The map's own `.pairs`, opened for its bytes (`TaskInputs.mapSource`).
*
* Through `openSource`, the road a dropped file takes, because the map's source
* is exactly that: a `SourceSpec` the reader opened, local or remote. A venue
* that cannot open one says so rather than reading nothing.
*/
async function openOwn(job: JobContext, spec: SourceSpec, where: string): Promise<ByteSource | null> {
if (job.openSource === undefined) {
job.note(`Reading ${where} again needs a venue that can open a file it was handed, and this one cannot.`)
return null
}
try {
return await job.openSource(spec)
} catch (error) {
job.note(`Could not open ${where}: ${error instanceof Error ? error.message : String(error)}`)
return null
}
}
/**
* The bytes of a `.pairs` a run was given, by whichever road reaches them.
*
* A file **handed** to the form comes first — dropped, chosen, or picked out of a
* connected place — because its source is exactly what the reader gave and the
* name beside it is only the record of it. An address is fetched as it stands,
* and a path is read where the venue has a filesystem, which is a shell's whole
* case and a worker's impossible one: each venue says which it is by the
* capability it answers (`JobContext.openFile`).
*
* Shared with P(s), whose retired `pairsUri` (`run/expected.ts`) is read the
* same way, so the two tasks cannot come to disagree about which file a name
* means.
*/
export async function openRecords(
job: JobContext,
uri: string,
handed: SourceSpec | undefined,
): Promise<ByteSource | null> {
const remote = /^(https?|s3):\/\//i.test(uri)
const open =
handed !== undefined && job.openSource !== undefined
? () => job.openSource!(handed)
: remote && job.openBytes !== undefined
? () => job.openBytes!(uri)
: !remote && job.openFile !== undefined
? () => job.openFile!(uri)
: null
if (open === null) {
job.note(
remote
? `Reading ${uri} needs a venue that can fetch an address, and this one reads local ` +
'files only. Fetch it first and name the file, or run it in the app.'
: `“${uri}” is not an address this can fetch, and the file is no longer in hand — a ` +
'dropped file is kept for the session it was dropped in. Drop it on the form again, ' +
'or name it on a `chromosomix` command line, which reads it from disk.',
)
return null
}
try {
return await open()
} catch (error) {
job.note(`Could not open ${uri}: ${error instanceof Error ? error.message : String(error)}`)
return null
}
}
// ---------------------------------------------------------------------------
// P(s) by orientation
// ---------------------------------------------------------------------------
/**
* The four orientation curves as a figure — or `null`, with a sentence, where
* fewer than two have anything — and, where `converge` asks for it, the figure's
* own sentence about where they meet.
*
* One function for both tasks that draw it: P(s) for a stored run that still
* asks (`run/expected.ts`), and Pairs QC, which is where the figure lives now.
* Re-binned at `factor` (P(s)'s own `factor` where it has one, 1.1 here),
* normalised over the records in all four (`orientationFigure`).
*
* **Pairs QC states no convergence of the figure's own.** The figure's rule —
* the four within a tenth in log space, walking down from the long end — and
* pairtools' — any of them five per cent off the four's mean, the furthest such
* bin — are two answers to one question, and the task that also prints
* `pairtools stats` must give one: so it passes `converge: false` and says
* pairtools' instead ({@link convergenceNote}). P(s)'s retired curve keeps the
* sentence it always had.
*/
export function orientationOutput(
job: JobContext,
counts: OrientationCounts,
id: string,
options: { factor?: number; converge?: boolean } = {},
): CurveOutput | null {
const factor = options.factor ?? Math.max(1.01, numParam(job.params, 'factor', 1.1))
const figure = orientationFigure(counts, { factor })
if (figure === null) {
job.note(
'Fewer than two orientations have any cis record in them, so there is nothing to compare. ' +
'A file whose strand columns are all `.` reads this way.',
)
return null
}
if (options.converge !== false) job.note(
figure.convergence === null
? 'The four orientations never settle to within a tenth of each other, so no convergence ' +
'point is reported. On a library that is otherwise fine this usually means the axis does ' +
'not reach far enough — the sequences are short — rather than that the records are bad.'
: `The four orientations converge at about ${Math.round(figure.convergence).toLocaleString()} bp, ` +
'which is the shortest separation this library can be read at. Below it the counts are ' +
'self-circles, dangling ends and undigested fragments rather than contacts.',
)
return {
kind: 'curve',
id,
label: 'P(s) by pair orientation',
series: figure.curves.map(
(curve): CurveSeries => ({ name: curve.name, x: curve.x, y: curve.y, n: curve.n }),
),
xLabel: 'Genomic separation (bp)',
yLabel: 'Density, over the records in all four',
xScale: 'log',
yScale: 'log',
}
}
// ---------------------------------------------------------------------------
// pairtools stats
// ---------------------------------------------------------------------------
/**
* Where the four orientations meet, by pairtools' rule (`convergenceOf`): the
* edge after the furthest separation bin at which any orientation strays more
* than five per cent from the four's mean — `summary/dist_freq_convergence/
* convergence_dist` in the table below, to the base pair.
*/
export function convergenceNote(stats: PairStats): string {
const found = convergenceOf(stats)
const share = (count: number): string =>
stats.cis > 0 ? `${((100 * count) / stats.cis).toFixed(1)} %` : '—'
if (found.distance === null) {
return (
'No orientation strays more than 5 % from the mean of the four at any separation, so ' +
'pairtools reports no convergence distance: the four agree from the first base pair.'
)
}
// The caveat is pairtools' own rule read honestly: the *furthest* bin with a
// stray sets the distance, so a sparse tail of chance strays can put it far
// past where the curves lie together, as on a small library it does.
return (
`The four orientations converge at ${found.distance.toLocaleString('en-US')} bp by ` +
`pairtools' rule: past the last separation bin at which one of them ` +
`(${found.strands ?? '?'}) strays more than 5 % from the mean of the four, with ` +
`${share(found.belowAll)} of the cis pairs below it. The furthest such bin sets it, so a ` +
'sparse tail of chance strays can put it well past where the curves lie together.'
)
}
/** Bars over a list, at `0, 1, 2…`, named by `ticks` — one figure's whole abscissa. */
function listAxis(ticks: readonly string[]): Float64Array {
return Float64Array.from(ticks, (_, at) => at)
}
/** Every pair type and how many records had it, most common first; `null` without the column. */
function typesOutput(stats: PairStats): CurveOutput | null {
if (stats.pairTypes.length === 0) return null
const ticks = stats.pairTypes.map(([type]) => type)
return {
kind: 'curve',
id: 'pairsqc-types',
label: 'Pair types',
series: [
{
name: 'records',
x: listAxis(ticks),
y: Float64Array.from(stats.pairTypes, ([, count]) => count),
mark: 'bars',
n: stats.total,
},
],
xLabel: 'Pair type',
yLabel: 'Records',
xScale: 'linear',
yScale: 'linear',
xTicks: ticks,
}
}
/**
* Cis and trans, and cis at least 1, 2, 4, 10, 20 and 40 kb apart, as shares
* of the pairs mapped on both sides and not duplicated — pairtools'
* `summary/frac_cis*` and its denominator, with trans beside cis.
*
* Named on the axis by pairtools' own keys, so a bar reads as the line of a
* `.stats` file it was drawn from.
*/
function cisOutput(stats: PairStats): CurveOutput {
const ticks = ['cis', 'trans', ...stats.cisAtLeast.map(([kb]) => `cis_${kb}kb+`)]
const counts = [stats.cis, stats.trans, ...stats.cisAtLeast.map(([, count]) => count)]
const share = (count: number): number => (stats.totalNodups > 0 ? count / stats.totalNodups : Number.NaN)
return {
kind: 'curve',
id: 'pairsqc-cis',
label: 'Cis and trans',
series: [
{
name: 'fraction',
x: listAxis(ticks),
y: Float64Array.from(counts, share),
mark: 'bars',
n: stats.totalNodups,
},
],
xLabel: 'Pairs',
yLabel: 'Share of the mapped pairs that are not duplicates',
xScale: 'linear',
yScale: 'linear',
// The whole of a fraction, for the reason the cis/trans task gives: a fitted
// axis draws two libraries that agree as two that do not.
yDomain: [0, 1],
xTicks: ticks,
}
}
/**
* The summary `pairtools stats` writes, as a table: every entry but the three
* long lists — chromosome pairs, the separation histogram and the chromsizes —
* which the figures above draw or which are the header's. Each value spelled as
* pairtools spells it (`statText`), so a row can be laid beside the line of a
* `.stats` file and compared character for character.
*/
function statsOutput(stats: PairStats): TableOutput {
const rows = pairtoolsStatsEntries(stats)
.filter(([key]) => !/^(chrom_freq|dist_freq|chromsizes)\//.test(key))
.map(([key, value]) => ({ statistic: key, value: statText(value) }))
return {
kind: 'table',
id: 'pairsqc-stats',
label: 'pairtools stats',
columns: [
{ key: 'statistic', label: 'Statistic', kind: 'text' },
{ key: 'value', label: 'Value', kind: 'text' },
],
rows,
}
}
// ---------------------------------------------------------------------------
// Restriction fragments
// ---------------------------------------------------------------------------
/** The four restriction outputs, and the sentence about what was counted. */
function restrictionOutputs(job: JobContext, stats: RestrictionStats): TaskOutput[] {
const enzymes = stats.enzymes === '' ? 'the enzyme' : stats.enzymes
job.note(
`${stats.counted.toLocaleString('en-US')} pairs were placed in ${enzymes}'s ` +
`${stats.fragments.total.toLocaleString('en-US')} fragments` +
(stats.unassigned > 0
? `; ${stats.unassigned.toLocaleString('en-US')} more had an end on a sequence the ` +
'reference has not got, and are not classified.'
: '.'),
)
return [
classesOutput(stats, enzymes),
classesTable(stats, enzymes),
distanceOutput(stats, enzymes),
fragmentsOutput(stats, enzymes),
]
}
/** Each class, stacked by the strands of its two ends. */
function classesOutput(stats: RestrictionStats, enzymes: string): CurveOutput {
const ticks = [...PAIR_CLASSES]
const x = listAxis(ticks)
return {
kind: 'curve',
id: 'pairsqc-classes',
label: `Pair classes by orientation · ${enzymes}`,
series: ORIENTATIONS.map(
(orientation): CurveSeries => ({
name: orientation,
x,
y: Float64Array.from(PAIR_CLASSES, (name) => stats.byOrientation[name][orientation]),
mark: 'bars',
stack: 'orientation',
}),
),
xLabel: 'Class',
yLabel: 'Pairs',
xScale: 'linear',
yScale: 'linear',
xTicks: ticks,
}
}
/** The same counts as a table, with each class's share of the pairs placed. */
function classesTable(stats: RestrictionStats, enzymes: string): TableOutput {
return {
kind: 'table',
id: 'pairsqc-classes-table',
label: `Pair classes · ${enzymes}`,
columns: [
{ key: 'class', label: 'Class', kind: 'text' },
{ key: 'pairs', label: 'Pairs', kind: 'number' },
{ key: 'share', label: 'Share', kind: 'number', digits: 4 },
...ORIENTATIONS.map((orientation) => ({
key: orientation,
label: orientation,
kind: 'number' as const,
})),
],
rows: PAIR_CLASSES.map((name) => ({
class: name,
pairs: stats.classes[name],
share: stats.counted > 0 ? stats.classes[name] / stats.counted : Number.NaN,
...stats.byOrientation[name],
})),
}
}
/**
* How far each end was from the next cut in the direction it reads, by strand.
*
* At each bin's **left edge**, as pairtools names its separation bins, so a row
* of the table is a bin by the number that starts it; and only from the first
* bin either strand has anything in to the last, since the empty decades either
* side are not a distance anybody measured. The first bin, `[0, 1)`, is never
* drawn: a distance is at least one base.
*/
function distanceOutput(stats: RestrictionStats, enzymes: string): CurveOutput {
const edges = stats.distanceEdges
const plus = stats.distance['+']
const minus = stats.distance['-']
let first = edges.length
let last = -1
for (let i = 1; i < edges.length; i++) {
if (plus[i] > 0 || minus[i] > 0) {
first = Math.min(first, i)
last = i
}
}
const at = last < first ? [] : Array.from({ length: last - first + 1 }, (_, k) => first + k)
const x = Float64Array.from(at, (i) => edges[i])
return {
kind: 'curve',
id: 'pairsqc-distance',
label: `Distance from a read to the next ${enzymes} cut`,
series: [
{ name: '+', x, y: Float64Array.from(at, (i) => plus[i]) },
{ name: '-', x, y: Float64Array.from(at, (i) => minus[i]) },
],
xLabel: 'Distance to the cut (bp, bin start)',
yLabel: 'Read ends',
xScale: 'log',
yScale: 'linear',
}
}
/** How many ends each fragment holds, with how many of them held any in the label. */
function fragmentsOutput(stats: RestrictionStats, enzymes: string): CurveOutput {
const { total, seen, binLabels, counts } = stats.fragments
return {
kind: 'curve',
id: 'pairsqc-fragments',
label:
`Read ends per ${enzymes} fragment · ${seen.toLocaleString('en-US')} / ` +
`${total.toLocaleString('en-US')} fragments seen`,
series: [
{
name: 'fragments',
x: listAxis(binLabels),
y: Float64Array.from(counts),
mark: 'bars',
n: total,
},
],
xLabel: 'Read ends in the fragment',
yLabel: 'Fragments',
xScale: 'linear',
yScale: 'linear',
xTicks: [...binLabels],
}
}
// ---------------------------------------------------------------------------
// Coverage against sites
// ---------------------------------------------------------------------------
/**
* The fewest bins a number of sites must hold for the mean line to be drawn
* there: a mean over one or two bins is those bins, and five is the fewest that
* read as a trend rather than as the bins themselves.
*/
const MEAN_MIN_BINS = 5
/** {@link MEAN_MIN_BINS} as the note says it. */
const MEAN_MIN_BINS_WORD = 'five'
/**
* Each bin's coverage against the cuts in it, with the mean coverage at each
* number of cuts, and Spearman's ρ between the two.
*
* **Coverage is the Coverage task's own** (`analysis/coverage.ts`): the raw
* contacts of every pixel touching the bin, cis and trans, its own diagonal
* pixel counted once — `cooltools.coverage(ignore_diags=0)`'s total less that
* pixel, which cooltools adds at both its ends. A bin's sites are the cuts `c`
* with `start <= c < end`, on the map's grid as it is read.
*
* Only the bins the reference speaks for: a sequence it has not got has no
* sites anybody counted, and its bins would read as zero. Of the rest, a bin
* with neither a contact nor a cut is left out of the correlation **and of the
* mean line**, which average the same bins; and the mean is drawn only at a
* number of sites {@link MEAN_MIN_BINS} bins share.
*
* The points are **sorted by their number of sites**, the bins of one number in
* the order the genome holds them, so the mean shares their abscissa: one row
* of the table per bin, its sites, its coverage and the mean of the bins with
* as many sites, which is a table the two series can be read across.
*/
async function coverageAgainstSites(
job: JobContext,
cuts: Map<string, CutSites>,
enzymes: string,
): Promise<CurveOutput | null> {
const mc = job.mc
const contigs: Array<{ startBin: number; bins: number }> = []
for (let cid = 0; cid < mc.nContigs; cid++) {
const [from, to] = mc.contigExtent(cid)
contigs.push({ startBin: from, bins: Math.max(0, to - from) })
}
// The sites first, which cost nothing: which bins the reference speaks for,
// and how many cuts fall in each.
const cutNames = [...cuts.keys()]
const sites = new Float64Array(mc.nBins)
const spoken = new Uint8Array(mc.nBins)
for (let cid = 0; cid < mc.nContigs; cid++) {
const at = contigIndexOf(cutNames, mc.chromnames[cid])
if (at < 0) continue
const { startBin, bins } = contigs[cid]
spoken.fill(1, startBin, startBin + bins)
// A site counts once, at its top-strand cut, as the contract's `c` is.
for (const cut of cuts.get(cutNames[at])?.top ?? []) {
const bin = startBin + Math.floor(cut / mc.binsize)
if (bin < startBin + bins) sites[bin]++
}
}
job.progress(0, 1, 'reading pixels')
const accumulator = new CoverageAccumulator(mc.nBins, binContigMap(contigs, mc.nBins))
await mc.scanPixels({
onChunk: (chunk) => {
job.throwIfCancelled()
accumulator.addPixels(chunk.bin1, chunk.bin2, chunk.count, chunk.length, null)
},
onProgress: (done, total) => job.progress(done, total, 'reading pixels'),
})
job.throwIfCancelled()
const coverage = accumulator.finish().total
const kept: number[] = []
let empty = 0
for (let bin = 0; bin < mc.nBins; bin++) {
if (spoken[bin] === 0) continue
if (coverage[bin] === 0 && sites[bin] === 0) {
empty++
continue
}
kept.push(bin)
}
if (kept.length === 0) {
job.note('No bin of the map lies on a sequence the reference has, so coverage is not set against sites.')
return null
}
kept.sort((a, b) => sites[a] - sites[b] || a - b)
const x = Float64Array.from(kept, (bin) => sites[bin])
const y = Float64Array.from(kept, (bin) => coverage[bin])
// The mean at each number of sites, over exactly the bins plotted — and only
// where {@link MEAN_MIN_BINS} of them share it. A number of sites one or two
// bins hold is a repeat or an rDNA array as often as not, and a line through
// their twelve thousand contacts draws the eye to them rather than to the
// trend; their points stay, with the mean left `NaN` — a gap in the line and
// an empty cell in the table.
const mean = new Float64Array(kept.length)
for (let start = 0; start < kept.length; ) {
let end = start
let sum = 0
while (end < kept.length && x[end] === x[start]) sum += y[end++]
mean.fill(end - start >= MEAN_MIN_BINS ? sum / (end - start) : Number.NaN, start, end)
start = end
}
const rho = spearman(x, y)
const rhoText = Number.isFinite(rho) ? rho.toFixed(3) : 'undefined'
job.note(
`Spearman ρ = ${rhoText} between the ${enzymes} sites in a bin and its coverage, over ` +
`${kept.length.toLocaleString('en-US')} bins of ${mc.binsize.toLocaleString('en-US')} bp` +
(empty === 0
? ''
: `; ${empty.toLocaleString('en-US')} ${empty === 1 ? 'bin' : 'bins'} with neither ` +
`${empty === 1 ? 'was' : 'were'} left out`) +
`. The mean is drawn where at least ${MEAN_MIN_BINS_WORD} bins share a number of sites.`,
)
return {
kind: 'curve',
id: 'pairsqc-sites',
label: `Coverage against ${enzymes} sites · ρ = ${rhoText}`,
series: [
{ name: 'bins', x, y, mark: 'points', n: kept.length },
{ name: 'mean', x, y: mean },
],
xLabel: 'Restriction sites in the bin',
yLabel: 'Contacts (coverage)',
xScale: 'linear',
yScale: 'linear',
}
}
|