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 | 2x 2x 2x 2x 2x 2x 2x 104x 104x 104x 2x 2x 2x 2x 2x 16x 16x 16x 16x 17x 17x 16x 16x 16x 16x 16x 10x 6x 6x 16x 16x 16x 4x 12x 16x 16x 16x 16x 15x 15x 15x 15x 15x 15x 15x 15x 15x 15x 15x 15x 104x 104x 15x 15x 15x 15x 34x 34x 34x 15x 34x 34x 34x 34x 34x 250x 250x 34x 34x 34x 34x 34x 34x 34x 34x 34x 34x 34x 34x 34x 15x 15x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 5x 15x 34x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 24x 34x 15x 15x 15x 15x 15x 34x 34x 34x 34x 34x 34x 34x 34x 34x 15x 15x 15x 15x 15x 15x 15x 15x 16x 15x 15x 29x 15x 15x 15x 15x 9x 6x 2x 4x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 4x 4x 4x 4x 4x 4x 4x 4x 4x 4x 16x 2x 2x 2x 15x 15x 12x 12x 12x 12x 12x 12x 12x 12x 12x 12x 1x 1x 1x 1x 1x 1x 1x 1x 11x 12x 12x 12x 12x 12x 11x 11x 11x 11x 10x 11x 11x 11x 11x 12x 12x 12x 11x 12x 12x 12x 11x 11x 11x 11x 11x 24x 24x 24x 24x 24x 24x 11x 11x 12x 11x 11x 11x 11x 29x 23x 23x 11x 11x 11x 12x 8x 8x 8x 8x 8x 8x 8x 8x 8x 8x 12x 12x 12x 12x 12x 12x 12x 12x 4x 4x 4x 4x 4x 4x 4x 4x 4x 4x 4x 4x 4x 40x 40x 4x 4x 4x 4x 4x 3x 3x 30x 30x 4x 1x 1x 1x 1x 1x 1x 4x 4x 4x 4x 4x 4x 14x 10x 10x 10x 14x 14x 4x 40x 30x 30x 30x 10x 10x 10x 4x 4x 4x 4x 4x 4x 4x 4x 4x 15x 15x 15x 15x 3x 15x 15x 15x 15x 15x 15x 15x 15x 15x 15x 1x 1x 1x 2x 2x 2x 2x 10x 10x 10x 10x 10x 1x 1x 40x 40x 40x 40x 47x 47x 34x 47x 40x 40x 6x 6x 6x 6x 6x 6x 2x 2x 6x 6x 6x 6x 6x 6x 6x 6x 6x 2x 6x 4x 2x 2x 2x 6x 6x 2x 43x 43x 20x 20x 43x 78x 78x 78x 78x 43x 1x 1x 1x 20x 43x 1x 1x 1x 20x 20x 2x 2x 21x 21x 21x 21x 148x 148x 127x 21x 21x 21x 21x 21x 148x 148x 148x 148x 150x 152x 152x 152x 152x 152x 150x 148x 148x 21x 21x 21x 23x 23x 23x 23x 131x 131x 129x 129x 23x 23x 152x 152x 23x 23x 21x 148x 148x 148x 152x 152x 152x 152x 152x 148x 148x 21x 21x 21x 152x 152x 152x 152x 43522x 43522x 43522x 43522x 43522x 43522x 43522x 43522x 152x 152x 2x 2x 2x 2x 2x 2x 2x 2x 2x 4x 4x 2x 2x 4x 4x 4x 4x 2x 2x 2x 2x 4x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 34x 34x 34x | /**
* Signal by group, run — in the annotations worker, beside the tracks and sets.
*
* Three steps. The **groups** are made: from a feature set, its intervals under
* each label merged, and what the set does not cover as a group of its own; from
* a track, the bins of the assembly cut by the track's sign or its quantiles.
* What is compared is **one value per entry** by default — each merged run of a
* label, each stretch between them, each run of bins in one group — or, with a
* bin size, stretches of at most a bin. Every **track** is then read over
* every one, as its exact mean over the bases it has a value for, at its
* own resolution rather than a zoom level's. And the groups
* are **compared**: summarised, drawn as violins, and tested by ranks pairwise
* and together (`analysis/groupstats.ts`), every test's q-value adjusted with
* every other's in the run (Benjamini–Hochberg).
*
* The assembly is `inputs.contigLengths`, which is what the tracks and the sets
* are laid out against in this venue; a set's sequence that the assembly does
* not have is named in a note rather than read.
*/
import { FeatureBuilder } from '../../annotations/featureset'
import type { FeatureSet, Region, TrackSamples } from '../../annotations/types'
import { fdrCorrection } from '../chromosight'
import { coversNoMatrix } from '../cliCommand'
import { kruskalWallis, mannWhitney, quantileOf, sortedFinite, summarise } from '../groupstats'
import {
boolParam,
chosenInput,
idParam,
listParam,
missingInput,
numParam,
strParam,
type TrackJobContext,
type TrackTaskRunner,
} from '../job'
import { quantileEdges } from '../stats'
import { GROUP_BINSIZE } from '../tasks/groupsignal'
import type { CurveOutput, IntervalsOutput, TableOutput, TaskOutput } from '../types'
/** One stretch compared: where it is, and the group it is in. */
interface Stretch {
contig: string
start: number
end: number
group: number
/**
* An entry that runs on past the end of a circular sequence: it covers
* `[0, wrap)` as well as `[start, end)`, `end` being the sequence's end.
*/
wrap?: number
}
/** The bases a stretch covers, both sides of a circle's origin. */
function basesOf(stretch: Stretch): number {
return stretch.end - stretch.start + (stretch.wrap ?? 0)
}
/** The groups a run compares, and the stretches that make them up. */
interface Grouping {
names: string[]
colours: string[]
stretches: Stretch[]
/** What the groups are of, for a figure's title: `A/B compartments, by name`. */
of: string
/** The grouping written out, for a grouping made from a track. */
segments?: FeatureSet
}
/** A colour per group: Okabe–Ito, and the rest of the genome grey. */
const PALETTE = ['#0072b2', '#d55e00', '#009e73', '#cc79a7', '#e69f00', '#56b4e9', '#f0e442', '#000000']
const REST_COLOUR = '#999999'
const REST = 'rest of the genome'
/** The most values a violin is drawn from; beyond it they are thinned evenly by rank. */
const DRAWN = 20_000
export const runGroupSignal: TrackTaskRunner = async (job) => {
const tracks = listParam(job.params, 'tracks')
if (tracks.length === 0) {
job.note('No track was chosen. Load a bigWig, or run a task that makes one, and tick it under Tracks.')
return []
}
const contigs: Array<{ name: string; length: number }> = []
for (const [name, length] of Object.entries(job.inputs.contigLengths ?? {})) {
if (length > 0) contigs.push({ name, length })
}
if (contigs.length === 0) {
job.note('This session knows no sequence lengths, so there is no genome to cut into stretches. Load a chrom.sizes, or the map itself.')
return []
}
// The bin size asked for. 0, the default, is one value per entry; a track is
// still cut on a grid, the map's resolution where there is a map and 10 kb
// where there is not, as base composition reads its own.
const asked = numParam(job.params, 'resolution', 0)
const perEntry = !(asked > 0)
const binsize =
asked > 0
? Math.round(asked)
: job.framing.binsize > 0 && !coversNoMatrix(job.framing)
? job.framing.binsize
: GROUP_BINSIZE
const circular = new Set(job.framing.circular ?? [])
let grouping =
strParam(job.params, 'groupBy', 'features') === 'track'
? await fromTrack(job, contigs, binsize, perEntry, circular)
: fromFeatures(job, contigs, perEntry ? Number.POSITIVE_INFINITY : binsize, circular)
if (grouping === null) return []
const unit = perEntry ? { one: 'entry', many: 'entries' } : { one: 'stretch', many: 'stretches' }
grouping = withoutEmpty(grouping, job, unit.one)
if (grouping.stretches.length === 0) {
job.note(`No ${unit.one} of the genome fell in any group, so there is nothing to compare.`)
return []
}
const labelOf = (id: string): string => job.inputs.labels?.[id] ?? id
const k = grouping.names.length
const summaryRows: Array<Record<string, string | number>> = []
const testRows: Array<Record<string, string | number>> = []
const figures: CurveOutput[] = []
for (let t = 0; t < tracks.length; t++) {
job.throwIfCancelled()
const name = labelOf(tracks[t])
job.progress(t, tracks.length, name)
const values = await meansOver(job, tracks[t], grouping.stretches)
const byGroup: number[][] = grouping.names.map(() => [])
for (let i = 0; i < grouping.stretches.length; i++) {
if (Number.isFinite(values[i])) byGroup[grouping.stretches[i].group].push(values[i])
}
if (byGroup.every((group) => group.length === 0)) {
job.note(`${name} has no value over any stretch — its sequence names may not be the assembly's.`)
continue
}
let all = 0
let allSum = 0
for (const group of byGroup) {
for (const value of group) allSum += value
all += group.length
}
for (let g = 0; g < k; g++) {
const summary = summarise(byGroup[g])
const others = all - summary.n
const restMean = others > 0 ? (allSum - summary.mean * summary.n) / others : Number.NaN
let bases = 0
for (const stretch of grouping.stretches) {
if (stretch.group === g) bases += basesOf(stretch)
}
summaryRows.push({
track: name,
group: grouping.names[g],
n: summary.n,
bases,
mean: summary.mean,
sd: summary.sd,
median: summary.median,
q1: summary.q1,
q3: summary.q3,
vsRest: summary.n > 0 ? summary.mean - restMean : Number.NaN,
})
}
const kept = byGroup.map((group, g) => ({ group, g })).filter((entry) => entry.group.length > 0)
if (kept.length > 2) {
const test = kruskalWallis(kept.map((entry) => entry.group))
testRows.push({
track: name,
test: 'Kruskal–Wallis H',
a: `all ${kept.length} groups`,
b: '',
statistic: test.h,
effect: Number.NaN,
difference: Number.NaN,
p: test.p,
})
}
for (let i = 0; i < kept.length; i++) {
for (let j = i + 1; j < kept.length; j++) {
const a = kept[i]
const b = kept[j]
const test = mannWhitney(a.group, b.group)
testRows.push({
track: name,
test: 'Mann–Whitney U',
a: grouping.names[a.g],
b: grouping.names[b.g],
statistic: test.u,
effect: test.effect,
difference:
quantileOf(sortedFinite(a.group), 0.5) - quantileOf(sortedFinite(b.group), 0.5),
p: test.p,
})
}
}
figures.push({
kind: 'curve',
id: `groupsignal-${t + 1}`,
label: `${name} by ${grouping.of}`,
series: byGroup.map((group, g) => {
const drawn = thinned(group)
return {
name: grouping.names[g],
x: Float64Array.from(drawn, (_, i) => i),
y: drawn,
colour: grouping.colours[g],
n: group.length,
mark: 'violin' as const,
}
}),
xLabel: grouping.of,
yLabel: name,
xScale: 'linear',
yScale: 'linear',
})
}
job.progress(tracks.length, tracks.length, 'done')
if (figures.length === 0) return []
const q = fdrCorrection(testRows.map((row) => Number(row.p)))
testRows.forEach((row, i) => {
row.q = q[i]
})
const drawnFrom = figures.some((figure) => figure.series.some((series) => (series.n ?? 0) > DRAWN))
const count = grouping.stretches.length.toLocaleString()
const compared = !perEntry
? `${count} stretches of at most ${binsize.toLocaleString()} bp`
: grouping.segments !== undefined
? `${count} entries, each a run of ${binsize.toLocaleString()} bp bins in one group,`
: `${count} entries, each a run of features under one name or a stretch between them,`
job.note(
`Compared over ${compared} in ${k} groups, each the track's mean over the bases it has a ` +
`value for.${drawnFrom ? ` A violin of more than ${DRAWN.toLocaleString()} is drawn from that many, evenly by rank; the tests use every one.` : ''}`,
)
const outputs: TaskOutput[] = [...figures]
outputs.push({
kind: 'table',
id: 'groupsignal-groups',
label: `Groups · ${grouping.of}`,
columns: [
{ key: 'track', label: 'Track', kind: 'text' },
{ key: 'group', label: 'Group', kind: 'text' },
{ key: 'n', label: perEntry ? 'Entries' : 'Stretches', kind: 'number' },
{ key: 'bases', label: 'Bases', kind: 'number' },
{ key: 'mean', label: 'Mean', kind: 'number', digits: 4 },
{ key: 'sd', label: 'SD', kind: 'number', digits: 4 },
{ key: 'median', label: 'Median', kind: 'number', digits: 4 },
{ key: 'q1', label: 'Q1', kind: 'number', digits: 4 },
{ key: 'q3', label: 'Q3', kind: 'number', digits: 4 },
{
key: 'vsRest',
label: 'Mean − rest',
kind: 'number',
digits: 4,
hint: 'This group’s mean less that of every other stretch.',
},
],
rows: summaryRows,
} satisfies TableOutput)
outputs.push({
kind: 'table',
id: 'groupsignal-tests',
label: 'Between groups',
columns: [
{ key: 'track', label: 'Track', kind: 'text' },
{ key: 'test', label: 'Test', kind: 'text' },
{ key: 'a', label: 'Group', kind: 'text' },
{ key: 'b', label: 'Against', kind: 'text' },
{ key: 'difference', label: 'Median difference', kind: 'number', digits: 4 },
{
key: 'effect',
label: 'Effect',
kind: 'number',
digits: 3,
hint: 'Rank-biserial r: +1 where every value of the group is above every value against it.',
},
{ key: 'statistic', label: 'U or H', kind: 'number', digits: 2 },
{ key: 'p', label: 'P', kind: 'number', digits: 4, scientific: true },
{
key: 'q',
label: 'q',
kind: 'number',
digits: 4,
scientific: true,
hint: 'Benjamini–Hochberg, over every test of the run.',
},
],
rows: testRows,
} satisfies TableOutput)
if (grouping.segments !== undefined) {
outputs.push({
kind: 'intervals',
id: 'groupsignal-segments',
label: `Groups · ${grouping.of}`,
payload: { kind: 'features', data: grouping.segments },
noun: { one: 'segment', many: 'segments' },
display: 'lane',
look: 'corner',
} satisfies IntervalsOutput)
}
if (strParam(job.params, 'groupBy', 'features') !== 'track' && boolParam(job.params, 'perFeature', false)) {
const perFeature = await perFeatureTable(job, tracks, labelOf)
if (perFeature !== null) outputs.push(perFeature)
}
return outputs
}
// ---------------------------------------------------------------------------
// The groups
// ---------------------------------------------------------------------------
/**
* Groups from a feature set: what is written on each feature, and the rest. A
* `binsize` of infinity cuts nothing, so each merged run is one entry.
*/
function fromFeatures(
job: TrackJobContext,
contigs: ReadonlyArray<{ name: string; length: number }>,
binsize: number,
circular: ReadonlySet<string>,
): Grouping | null {
const offered = chosenInput(job.inputs.features, job.params, 'features')
// Only a set the form chose: an input nobody picked is not one to group by.
const found = { chosen: offered.chosen, value: offered.chosen ? offered.value : undefined }
const set = found.value
if (set === undefined) {
job.note(
missingInput(found, {
unchosen: 'No feature set was chosen to make the groups from.',
gone: 'The feature set chosen is no longer loaded. Re-open the file, or pick another.',
}) ?? '',
)
return null
}
const split = strParam(job.params, 'split', 'name')
const setName = job.inputs.labels?.[idParam(job.params, 'features') ?? ''] ?? 'the set'
// A label per feature, and the order the groups are listed in.
const labels: Array<string | null> = new Array(set.length).fill(null)
let order: string[]
if (split === 'score') {
const k = quantileCount(job)
const edges = quantileEdges(set.score, k)
if (edges.length === 0) {
job.note(`${setName} carries no scores, so it cannot be cut into quantiles of them.`)
return null
}
order = Array.from({ length: k }, (_, i) => quantileName(i, edges))
for (let i = 0; i < set.length; i++) {
const at = bucketOf(set.score[i], edges)
if (at >= 0) labels[i] = order[at]
}
} else if (split === 'strand') {
order = ['+ strand', '− strand', 'no strand']
for (let i = 0; i < set.length; i++) {
labels[i] = set.strand[i] > 0 ? order[0] : set.strand[i] < 0 ? order[1] : order[2]
}
} else if (split === 'none') {
order = [setName]
labels.fill(setName)
} else {
for (let i = 0; i < set.length; i++) labels[i] = set.name[i] === '' ? '(no name)' : set.name[i]
order = [...new Set(labels.filter((label): label is string => label !== null))].sort((a, b) =>
a.localeCompare(b, undefined, { numeric: true }),
)
}
const present = order.filter((label) => labels.includes(label))
const withRest = boolParam(job.params, 'outside', true)
const names = withRest ? [...present, REST] : present
const index = new Map(names.map((name, i) => [name, i]))
// Each group's intervals per sequence, merged, and every feature's run for the rest.
const lengthOf = (contig: string): number | undefined =>
contigs.find((entry) => entry.name === contig)?.length ?? job.lengthOf(contig)
const runs = new Map<string, Array<Array<[number, number]>>>()
const unknown = new Set<string>()
for (let cid = 0; cid < set.contigs.length; cid++) {
const contig = set.contigs[cid]
const length = lengthOf(contig)
if (length === undefined) {
if (set.contigAt[cid + 1] > set.contigAt[cid]) unknown.add(contig)
continue
}
const perGroup: Array<Array<[number, number]>> = names.map(() => [])
for (let i = set.contigAt[cid]; i < set.contigAt[cid + 1]; i++) {
const label = labels[i]
if (label === null) continue
const start = Math.max(0, set.start[i])
const end = Math.min(length, set.end[i])
if (end > start) perGroup[index.get(label) ?? 0].push([start, end])
}
runs.set(contig, perGroup.map(merged))
}
if (unknown.size > 0) {
job.note(
`${[...unknown].slice(0, 6).join(', ')}${unknown.size > 6 ? ', …' : ''} ` +
`${unknown.size === 1 ? 'is' : 'are'} not a sequence of this assembly, so ${unknown.size === 1 ? 'its' : 'their'} features were left out.`,
)
}
const stretches: Stretch[] = []
const covered = new Map<string, Array<[number, number]>>()
for (const [contig, perGroup] of runs) {
perGroup.forEach((intervals, g) => {
for (const [start, end] of intervals) {
for (const [from, to] of cut(start, end, binsize)) stretches.push({ contig, start: from, end: to, group: g })
}
})
covered.set(contig, merged(perGroup.flat()))
}
if (withRest) {
const rest = names.length - 1
for (const { name, length } of contigs) {
let from = 0
for (const [start, end] of covered.get(name) ?? []) {
for (const [a, b] of cut(from, start, binsize)) stretches.push({ contig: name, start: a, end: b, group: rest })
from = Math.max(from, end)
}
for (const [a, b] of cut(from, length, binsize)) stretches.push({ contig: name, start: a, end: b, group: rest })
}
}
const splitWord = split === 'score' ? 'score' : split === 'strand' ? 'strand' : split === 'none' ? '' : 'name'
return {
names,
colours: names.map((name, i) => (name === REST ? REST_COLOUR : PALETTE[i % PALETTE.length])),
stretches: binsize === Number.POSITIVE_INFINITY ? acrossOrigin(stretches, contigs, circular) : stretches,
of: splitWord === '' ? setName : `${setName}, by ${splitWord}`,
}
}
/**
* Groups from a track: the assembly's bins, by the track's sign or its
* quantiles — compared a bin at a time, or `perEntry` a run of them at a time.
*/
async function fromTrack(
job: TrackJobContext,
contigs: ReadonlyArray<{ name: string; length: number }>,
binsize: number,
perEntry: boolean,
circular: ReadonlySet<string>,
): Promise<Grouping | null> {
const entry = idParam(job.params, 'groupTrack')
if (entry === null) {
job.note('No track was chosen to group the bins by.')
return null
}
const name = job.inputs.labels?.[entry] ?? entry
const bins: Stretch[] = []
for (const contig of contigs) {
for (let start = 0; start < contig.length; start += binsize) {
bins.push({ contig: contig.name, start, end: Math.min(contig.length, start + binsize), group: -1 })
}
}
const values = await meansOver(job, entry, bins)
const cutBy = strParam(job.params, 'cut', 'sign') === 'quantiles' ? 'quantiles' : 'sign'
let names: string[]
if (cutBy === 'sign') {
names = ['above 0', 'below 0']
for (let i = 0; i < bins.length; i++) {
bins[i].group = values[i] > 0 ? 0 : values[i] < 0 ? 1 : -1
}
} else {
const k = quantileCount(job)
const edges = quantileEdges(values, k)
if (edges.length === 0) {
job.note(`${name} has no value over any bin, so there is nothing to cut into quantiles.`)
return null
}
names = Array.from({ length: k }, (_, i) => quantileName(i, edges))
for (let i = 0; i < bins.length; i++) bins[i].group = bucketOf(values[i], edges)
}
const stretches = bins.filter((bin) => bin.group >= 0)
if (stretches.length < bins.length) {
job.note(
`${(bins.length - stretches.length).toLocaleString()} of ${bins.length.toLocaleString()} bins ` +
`had no reading of ${name}${cutBy === 'sign' ? ', or read exactly 0,' : ''} and are in no group.`,
)
}
// The grouping as intervals: consecutive bins of one group on one sequence, joined.
const segments = new FeatureBuilder()
const runs: Stretch[] = []
let open: Stretch | null = null
const close = (): void => {
if (open !== null) {
segments.add(open.contig, open.start, open.end, { name: names[open.group] })
runs.push(open)
}
open = null
}
for (const bin of stretches) {
if (open !== null && open.contig === bin.contig && open.group === bin.group && open.end === bin.start) {
open.end = bin.end
continue
}
close()
open = { ...bin }
}
close()
return {
names,
colours: names.map((_, i) => PALETTE[i % PALETTE.length]),
stretches: perEntry ? acrossOrigin(runs, contigs, circular) : stretches,
of: `${name}, ${cutBy === 'sign' ? 'by sign' : 'in quantiles'}`,
segments: segments.finish(),
}
}
/**
* The groups that have a stretch in them, and a note naming those that do not:
* the rest of the genome, where a compartment call covers every base, or a
* quantile that ties left empty. A violin of nothing and a row of dashes say
* less than the sentence does.
*/
function withoutEmpty(grouping: Grouping, job: TrackJobContext, unit: string): Grouping {
const counts = new Array<number>(grouping.names.length).fill(0)
for (const stretch of grouping.stretches) counts[stretch.group]++
if (counts.every((count) => count > 0)) return grouping
const empty = grouping.names.filter((_, g) => counts[g] === 0)
job.note(`${empty.join(', ')} had no ${unit} in ${empty.length === 1 ? 'it' : 'them'}, and ${empty.length === 1 ? 'is' : 'are'} left out.`)
const kept = grouping.names.map((_, g) => g).filter((g) => counts[g] > 0)
const renumbered = new Map(kept.map((g, i) => [g, i]))
return {
...grouping,
names: kept.map((g) => grouping.names[g]),
colours: kept.map((g) => grouping.colours[g]),
stretches: grouping.stretches.map((stretch) => ({ ...stretch, group: renumbered.get(stretch.group) ?? 0 })),
}
}
function quantileCount(job: TrackJobContext): number {
return Math.max(2, Math.min(10, Math.round(numParam(job.params, 'quantiles', 4))))
}
/** `Q2 (0.412 to 0.587)`: which quantile, and the values it spans. */
function quantileName(i: number, edges: Float64Array): string {
const format = (value: number): string => (Math.abs(value) >= 100 ? value.toFixed(0) : value.toPrecision(3))
return `Q${i + 1} (${format(edges[i])} to ${format(edges[i + 1])})`
}
/** Which quantile a value falls in, the last edge inclusive; −1 for none. */
function bucketOf(value: number, edges: Float64Array): number {
if (!Number.isFinite(value) || edges.length < 2) return -1
const k = edges.length - 1
if (value < edges[0] || value > edges[k]) return -1
for (let i = 0; i < k; i++) if (value < edges[i + 1]) return i
return k - 1
}
/** Intervals merged where they overlap or touch, in order. */
function merged(intervals: ReadonlyArray<[number, number]>): Array<[number, number]> {
const sorted = [...intervals].sort((a, b) => a[0] - b[0])
const out: Array<[number, number]> = []
for (const [start, end] of sorted) {
const last = out[out.length - 1]
if (last !== undefined && start <= last[1]) last[1] = Math.max(last[1], end)
else out.push([start, end])
}
return out
}
/**
* An entry that runs on past the end of a circular sequence, as one entry: the
* run of a group that ends at the sequence's end carries on over the run of the
* same group that starts at 0, rather than counting as two — a compartment
* across a bacterial chromosome's coordinate origin is one compartment.
*/
function acrossOrigin(
stretches: Stretch[],
contigs: ReadonlyArray<{ name: string; length: number }>,
circular: ReadonlySet<string>,
): Stretch[] {
if (circular.size === 0) return stretches
const lengths = new Map(contigs.map((contig) => [contig.name, contig.length]))
const ends = new Map<string, { head?: Stretch; tail?: Stretch }>()
for (const stretch of stretches) {
const length = lengths.get(stretch.contig)
if (!circular.has(stretch.contig) || length === undefined) continue
const key = `${stretch.contig}\t${stretch.group}`
const held = ends.get(key) ?? {}
if (stretch.start === 0 && stretch.end < length) held.head = stretch
if (stretch.end === length && stretch.start > 0) held.tail = stretch
ends.set(key, held)
}
const joined = new Set<Stretch>()
for (const { head, tail } of ends.values()) {
if (head === undefined || tail === undefined) continue
tail.wrap = head.end
joined.add(head)
}
return joined.size === 0 ? stretches : stretches.filter((stretch) => !joined.has(stretch))
}
/**
* A run cut into stretches of at most a bin, on the genome's bin grid.
*
* A run no longer than a bin is one stretch, whatever grid line it crosses — a
* peak is one observation. A longer one is cut where the grid falls, and a piece
* shorter than half a bin at either end is joined to its neighbour rather than
* kept as a sliver whose mean is a handful of bases.
*/
export function cut(start: number, end: number, binsize: number): Array<[number, number]> {
if (!(end > start)) return []
if (end - start <= binsize) return [[start, end]]
const pieces: Array<[number, number]> = []
let from = start
while (from < end) {
const to = Math.min(end, (Math.floor(from / binsize) + 1) * binsize)
pieces.push([from, to])
from = to
}
if (pieces.length > 1 && pieces[0][1] - pieces[0][0] < binsize / 2) {
pieces[1][0] = pieces[0][0]
pieces.shift()
}
const last = pieces.length - 1
if (last > 0 && pieces[last][1] - pieces[last][0] < binsize / 2) {
pieces[last - 1][1] = pieces[last][1]
pieces.pop()
}
return pieces
}
// ---------------------------------------------------------------------------
// Reading
// ---------------------------------------------------------------------------
/**
* How many bases one request asks for, at most, and how many stretches — a
* stretch longer than that, an entry a compartment or a chromosome long, is
* read a piece at a time and its pieces summed.
*
* Every track is read **at its own resolution** — a bigWig's base-level records,
* never a zoom level — because a stretch's value is the mean over the bases it
* covers, and a zoom record is a mean over *its* span: read at a twentieth of a
* stretch, a mouse ATAC-seq track's 896 bp records put the next stretch's signal
* into this one's ends, and chr19's 10 kb stretches came out a median 0.2 % from
* their own means, one in thirteen more than 1 %. Reading at base resolution
* makes the read the whole file, which is what an exact mean costs; a megabase at
* a time keeps what is held to one megabase's records, as track correlation
* reads its own (`run/trackcorr.ts`).
*/
const CHUNK_BASES = 1 << 20
const CHUNK_STRETCHES = 4096
/**
* A track's mean over each stretch — over the bases it has a value for, so a
* stretch half of which the track does not cover reports the half it does —
* each record weighted by how much of the stretch it covers: pyBigWig's
* `stats(…, exact=True)`.
*
* By sequence, the stretches in order along it and read a batch at a time, and
* matched back by the coordinates each reply echoes: `sample` drops what its
* file cannot answer, so matching by position would shift every one after it.
*/
async function meansOver(job: TrackJobContext, entry: string, stretches: readonly Stretch[]): Promise<Float64Array> {
const out = new Float64Array(stretches.length).fill(Number.NaN)
const byContig = new Map<string, number[]>()
stretches.forEach((stretch, i) => {
const held = byContig.get(stretch.contig)
if (held === undefined) byContig.set(stretch.contig, [i])
else held.push(i)
})
for (const [contig, indices] of byContig) {
// Each piece once, however many groups or features share it.
const keyed = new Map<string, Region>()
const piecesOf = new Map<number, string[]>()
for (const i of indices) {
const { start, end, wrap } = stretches[i]
const keys: string[] = []
const spans: Array<[number, number]> = wrap === undefined ? [[start, end]] : [[start, end], [0, wrap]]
for (const [from, to] of spans) {
for (let at = from; at < to; at += CHUNK_BASES) {
const piece = { contig, start: at, end: Math.min(to, at + CHUNK_BASES) }
const key = `${piece.start}-${piece.end}`
keyed.set(key, piece)
keys.push(key)
}
}
piecesOf.set(i, keys)
}
const regions = [...keyed.values()].sort((a, b) => a.start - b.start || a.end - b.end)
const sums = new Map<string, [number, number]>()
for (let first = 0; first < regions.length; ) {
job.throwIfCancelled()
let last = first + 1
let bases = regions[first].end - regions[first].start
while (last < regions.length && last - first < CHUNK_STRETCHES) {
bases += regions[last].end - regions[last].start
if (bases > CHUNK_BASES) break
last++
}
const sampled = await job.sample(entry, regions.slice(first, last), 1)
for (const answer of sampled.regions) {
sums.set(`${answer.start}-${answer.end}`, clippedSum(answer, answer.start, answer.end))
}
first = last
}
for (const i of indices) {
let sum = 0
let covered = 0
for (const key of piecesOf.get(i) ?? []) {
const held = sums.get(key)
if (held === undefined) continue
sum += held[0]
covered += held[1]
}
if (covered > 0) out[i] = sum / covered
}
}
return out
}
/** The sum of a track's values over `[start, end)`, each weighted by the bases it covers, and those bases. */
function clippedSum(answer: TrackSamples, start: number, end: number): [number, number] {
let sum = 0
let covered = 0
for (let k = 0; k < answer.values.length; k++) {
const value = answer.values[k]
if (!Number.isFinite(value)) continue
const from = Math.max(start, answer.starts[k])
const to = Math.min(end, answer.ends[k])
if (to <= from) continue
sum += value * (to - from)
covered += to - from
}
return [sum, covered]
}
/** Every feature of the set, unmerged, with its mean on every track. */
async function perFeatureTable(
job: TrackJobContext,
tracks: readonly string[],
labelOf: (id: string) => string,
): Promise<TableOutput | null> {
const set = chosenInput(job.inputs.features, job.params, 'features').value
if (set === undefined) return null
const features: Stretch[] = []
for (let i = 0; i < set.length; i++) {
features.push({ contig: set.contigs[set.contig[i]], start: set.start[i], end: set.end[i], group: i })
}
const columns = tracks.map((_, t) => `track${t + 1}`)
const rows = features.map((feature) => ({
contig: feature.contig,
start: feature.start,
end: feature.end,
name: set.name[feature.group] ?? '',
})) as Array<Record<string, string | number>>
for (let t = 0; t < tracks.length; t++) {
const values = await meansOver(job, tracks[t], features)
values.forEach((value, i) => {
rows[i][columns[t]] = value
})
}
return {
kind: 'table',
id: 'groupsignal-features',
label: 'Per feature',
columns: [
{ key: 'contig', label: 'Sequence', kind: 'text' },
{ key: 'start', label: 'Start', kind: 'number', coordinate: 'start' },
{ key: 'end', label: 'End', kind: 'number', coordinate: 'end' },
{ key: 'name', label: 'Name', kind: 'text' },
...tracks.map((id, t) => ({ key: columns[t], label: labelOf(id), kind: 'number' as const, digits: 4 })),
],
rows,
}
}
/** At most {@link DRAWN} of a group's values, evenly spaced by rank. */
function thinned(values: readonly number[]): Float64Array {
const sorted = sortedFinite(values)
if (sorted.length <= DRAWN) return sorted
const out = new Float64Array(DRAWN)
for (let i = 0; i < DRAWN; i++) out[i] = sorted[Math.round((i * (sorted.length - 1)) / (DRAWN - 1))]
return out
}
|