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 | 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 132x 132x 132x 396x 396x 396x 132x 2x 2x 482x 482x 2x 362x 362x 362x 2x 1475x 1475x 2x 1991x 1991x 2x 2x 41x 41x 41x 214x 214x 214x 214x 41x 41x 51x 51x 51x 51x 51x 9x 9x 9x 2x 7x 9x 9x 48x 1x 1x 41x 48x 48x 1x 1x 1x 1x 1x 39x 39x 2x 17x 17x 2x 17x 17x 2x 319x 319x 319x 262x 262x 227x 227x 1x 226x 262x 53x 317x 52x 52x 52x 50x 50x 48x 48x 50x 86x 60x 60x 1x 1x 1x 59x 60x 10x 10x 10x 49x 60x 38x 38x 293x 41x 41x 2x 49x 49x 2x 396x 396x 396x 396x 2x 1972x 1972x 1972x 1972x 24x 1972x 24x 24x 24x 24x 2x 1673x 1673x 1673x 1673x 1673x 1095572x 1095572x 1095553x 1095553x 1095553x 1673x 1673x 1673x 1673x 1673x 1673x 1095553x 1095553x 1095553x 1095553x 1095553x 1095553x 1673x 1673x 2x 1141x 1141x 1141x 1972x 1972x 1972x 7x 7x 7x 7x 1972x 1141x 1141x 1141x 1141x 2x 532x 532x 532x 532x 532x 532x 1138x 547785x 547785x 547785x 1138x 532x 532x 2x 54926x 54926x 54926x 149319874x 149319874x 149319874x 54926x 54926x 54926x 2x 117789x 117789x 117789x 1290029x 1290029x 1290029x 1290029x 117789x 117789x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 10x 10x 10x 10x 10x 10x 10x 10x 4x 4x 4x 10x 10x 2x 2x 2x 2x 2x 2x 39317x 39317x 13300x 13300x 13300x 39317x 39317x 2x 95924x 95924x 95924x 37251x 37251x 37251x 121x 121x 49x 49x 49x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 27x 27x 27x 27x 27x 27x 27x 44x 44x 2x 2x 2x 44x 43x 43x 43x 43x 43x 43x 43x 43x 27x 27x 27x 27x 27x 1769472x 1769472x 1769472x 27x 2x 228x 228x 10x 10x 10x 228x 2x 51524x 51524x 51524x 51524x 51524x 51524x 51524x 51524x 51524x 47962x 47962x 51524x 51524x 44041x 51524x 57x 57x 57x 57x 57x 44041x 51524x 171x 171x 171x 171x 171x 51524x 3009x 3009x 3009x 41032x 41032x 41032x 51524x 51524x 51524x 26520x 26520x 26520x 26520x 51524x 51524x 51524x 36648x 36648x 36648x 36648x 41032x 41032x 41032x 41032x 41032x 41032x 51524x 1715x 1715x 51524x 4781x 4781x 39317x 34536x 34536x 34536x 41032x 51524x 2x 82064x 82064x 82064x 82064x 40980x 82064x 2x 28x 28x 28x 168x 168x 168x 672x 672x 672x 168x 168x 168x 28x 28x 28x 28x 37251x 37251x 37251x 37251x 37251x 37251x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 28x 2x 2x 12x 12x 11x 12x 12x 11x 11x 11x 11x 12x 16x 16x 16x 16x 16x 16x 1334x 1334x 1334x 14x 14x 14x 14x 14x 1334x 1334x 1334x 16x 16x 11x 11x | /**
* Where a restriction enzyme cuts a reference, and what that says about a
* library of pairs.
*
* A Hi-C library is made by cutting chromatin with an enzyme and ligating the
* ends back together, so every read pair carries the trace of the cuts: the two
* ends sit near a site, a pair whose ends share a fragment was never ligated to
* anything else, and a fragment no pair touched is one the library never saw.
* pairtools answers those questions from a fragment BED written by
* `bioframe.digest` and `pairtools restrict`; this module is the same
* computation in one pass, beside a reference already loaded, with no file
* written in between.
*
* It is the whole of the computation and nothing else — no DOM, no worker, no
* store. `analysis/run/restriction.ts` walks a reference with {@link cutsIn},
* and the Pairs QC task feeds a `.pairs` through a {@link RestrictionCounter};
* both are wiring around what is here.
*
* ## The cut, exactly — on both strands
*
* A site is written with a `^` where the enzyme cuts the top strand —
* `^GATC`, `G^ANTC`, `CATG^` — and {@link Enzyme.cut} is the number of site
* bases `k` before it. A match whose first base is at 0-based `s` cuts the top
* strand at `c = s + k`, the index of the first base after the cut, and the
* bottom strand at `c' = s + L - k` in the same coordinates, `L` being the
* site's length: DpnII cuts at `s` and `s + 4`, HinfI at `s + 1` and `s + 4`,
* NlaIII at `s + 4` and `s`, AluI at `s + 2` twice. Every preset is a
* palindrome, so its two strands give one cut.
*
* A typed site that is not a palindrome also matches its reverse complement,
* and its bottom strand is taken to be cut symmetrically, as a palindrome's is.
* A reverse match at `s` then cuts at the same two places a forward one does —
* the site's strand `k` bases from its 5′ end, which is `s + L - k` on the
* bottom, and the other strand `k` bases from its own, `s + k` on the top — so
* the top cut is `s + k` whichever strand the site was read on. That is where
* Biopython puts it too: its `search` reports a reverse BssSI (`C^ACGAG`) or
* BbvCI (`CC^TCAGC`) site at the same top-strand cut as a forward one, and the
* golden holds a typed `C^ACGAG` to BssSI's digest.
*
* The matching is `annotations/motif.ts`' and is not repeated here: the IUPAC
* masks, the rule that a reference `N` matches only a pattern `N`, case
* ignored, every overlapping match found. What this adds is the offset from a
* match to its cuts.
*
* **`bioframe.digest` is two bases late**, and the goldens and the validation
* notebook allow for it rather than copy it: Biopython's `search` already
* returns the 1-based first base after the cut, and bioframe adds one more on
* the way to a 0-based start. So its boundaries are the top-strand cuts plus
* two, at every site (`scripts/make_restriction_golden.py`).
*
* ## The fragment, exactly — placed by the strand that was read
*
* On a sequence of length `N`, the top boundaries are `B = [0, …c, N]` and the
* bottom ones `B' = [0, …c', N]`, both in the order of `c`, so fragment `i` is
* one fragment: `[B[i], B[i+1])` on the top strand and `[B'[i], B'[i+1])` on
* the bottom. A pair end at the 1-based `pos` a `.pairs` file writes is at
* `x = pos - 1`; a `+` end lies in fragment `searchsorted(B, x, 'right') - 1`
* and a `-` end in `searchsorted(B', x, 'right') - 1`, each clamped to the
* sequence's fragments as `pairtools restrict` clamps (see {@link fragmentOf}).
*
* The bottom strand is the part that is easy to get wrong. A `-` read's 5′ end
* is the bottom strand's, and once a 5′ overhang is filled in, a fragment's end
* on the bottom strand is at `c' - 1` — three bases past the top-strand cut for
* DpnII. Placed by the top-strand cut, the `-` read of every DpnII dangling end
* falls in the next fragment and the pair reads as a re-ligation.
*/
import { compileMotif, findMotif, isMotifRefusal, type Motif } from '../annotations/motif'
import type { IntervalsOutput, TaskResult } from '../analysis/types'
import { distEdges } from '../pairs/stats'
import { contigIndexOf } from './naming'
// ---------------------------------------------------------------------------
// Enzymes
// ---------------------------------------------------------------------------
/** One enzyme: its site in upper-case IUPAC letters, and where in it the top strand is cut. */
export interface Enzyme {
name: string
site: string
/** Site bases before the `^`: `0` for `^GATC`, `1` for `G^ANTC`, `4` for `CATG^`. */
cut: number
}
/**
* The presets, in the order the form lists them.
*
* The four-cutters Hi-C protocols use and the six-cutters of the first Hi-C
* papers. Each site is checked against Biopython's `Bio.Restriction` by the
* golden script, which digests the yeast reference with every one of them and
* holds {@link cutsIn} to the answer.
*/
export const ENZYMES: readonly Enzyme[] = [
{ name: 'DpnII', site: 'GATC', cut: 0 },
{ name: 'HindIII', site: 'AAGCTT', cut: 1 },
{ name: 'NcoI', site: 'CCATGG', cut: 1 },
{ name: 'EcoRI', site: 'GAATTC', cut: 1 },
{ name: 'BglII', site: 'AGATCT', cut: 1 },
{ name: 'HinfI', site: 'GANTC', cut: 1 },
{ name: 'DdeI', site: 'CTNAG', cut: 1 },
{ name: 'MseI', site: 'TTAA', cut: 1 },
{ name: 'NlaIII', site: 'CATG', cut: 4 },
{ name: 'CviQI', site: 'GTAC', cut: 1 },
{ name: 'AluI', site: 'AGCT', cut: 2 },
]
/**
* Other names for a preset: isoschizomers, which recognise the same site and
* cut it in the same place, so a reader who used MboI gets DpnII's fragments
* under the name they typed.
*/
export const ENZYME_ALIASES: Readonly<Record<string, string>> = {
MboI: 'DpnII',
Sau3AI: 'DpnII',
Csp6I: 'CviQI',
}
/**
* Kits that cut with more than one enzyme. Micro-C and Omni-C cut with a
* nuclease rather than an enzyme and have no sites to list, so they are not
* here: a library made with one asks for no enzyme.
*/
export const KITS: Readonly<Record<string, readonly string[]>> = {
Arima: ['DpnII', 'HinfI'],
}
const BY_NAME = ((): Map<string, Enzyme> => {
const out = new Map<string, Enzyme>()
for (const enzyme of ENZYMES) out.set(enzyme.name.toLowerCase(), enzyme)
for (const [alias, name] of Object.entries(ENZYME_ALIASES)) {
const enzyme = out.get(name.toLowerCase())
if (enzyme !== undefined) out.set(alias.toLowerCase(), enzyme)
}
return out
})()
/** A preset by its name or an alias, case ignored; `undefined` for one this does not know. */
export function enzymeByName(name: string): Enzyme | undefined {
return BY_NAME.get(name.trim().toLowerCase())
}
/** A kit's name as {@link KITS} spells it, case ignored, or `undefined`. */
export function kitNamed(name: string): string | undefined {
const wanted = name.trim().toLowerCase()
return Object.keys(KITS).find((kit) => kit.toLowerCase() === wanted)
}
/** The site as it is written, with the `^`: `^GATC`, `G^ANTC`. */
export function siteText(enzyme: Enzyme): string {
return `${enzyme.site.slice(0, enzyme.cut)}^${enzyme.site.slice(enzyme.cut)}`
}
/**
* How far past the top-strand cut the bottom strand is cut, `L - 2k`: `4` for
* DpnII, `3` for HinfI, `-4` for NlaIII, `0` for a blunt cutter such as AluI.
*/
export function bottomShift(enzyme: Enzyme): number {
return enzyme.site.length - 2 * enzyme.cut
}
/**
* The fewest sites per base a typed site may expect: a four-base cutter's.
*
* A guard on the shape of the request rather than a limit of the scan. `^GA`
* cuts every sixteen bases or so, which on a mammalian reference is two hundred
* million cuts and a lane nobody could draw — and it is never what a reader of a
* Hi-C library meant. Every preset is at this bound or under it; a degenerate
* position counts for what it permits, so `G^ANTC` is as specific as `^GATC`.
*/
const MOST_SPECIFIC_FREQUENCY = 1 / 256
/** How many enzymes one digest may name: a cut keeps which of them made it in a byte. */
const MOST_ENZYMES = 255
/** How often a compiled site is expected at a position of a uniform genome. */
function frequencyOf(motif: Motif): number {
let frequency = 1
for (const mask of motif) {
let bases = 0
for (let bit = mask; bit !== 0; bit &= bit - 1) bases++
frequency *= bases / 4
}
return frequency
}
/** One typed site, or why it cannot be one. */
function parseSite(text: string): Enzyme | { refusal: string } {
// `U` is `T`, as the matcher reads it, so the site is written in one alphabet.
const upper = text.toUpperCase().replace(/U/g, 'T')
const at = upper.indexOf('^')
const letters = upper.replace('^', '')
if (at < 0) {
const compiled = compileMotif(letters)
return {
refusal: isMotifRefusal(compiled)
? `“${text}” is neither an enzyme this knows nor a site in IUPAC letters with a ^ where ` +
'it cuts.'
: `“${text}” needs a ^ where the enzyme cuts the top strand, as in ^GATC or G^ANTC.`,
}
}
if (letters.includes('^')) {
return { refusal: `“${text}” has more than one ^, and a site is cut in one place.` }
}
const compiled = compileMotif(letters)
if (isMotifRefusal(compiled)) return { refusal: compiled.reason }
if (frequencyOf(compiled) > MOST_SPECIFIC_FREQUENCY) {
return {
refusal:
`“${text}” would cut every ${Math.round(1 / frequencyOf(compiled))} bases or so; a site ` +
'needs at least the specificity of a four-base cutter.',
}
}
return { name: `${upper.slice(0, at)}^${upper.slice(at + 1)}`, site: letters, cut: at }
}
/**
* The enzyme a name stands for — a preset, an alias, or a typed site written
* with its `^` — or `undefined`. How {@link cutsOfResult} reads the name each
* cut of a run carries.
*/
export function enzymeNamed(name: string): Enzyme | undefined {
const preset = enzymeByName(name)
if (preset !== undefined) return preset
const site = parseSite(name)
return 'refusal' in site ? undefined : site
}
/**
* The enzymes a form's two answers name, or why they name none that can run.
*
* `choice` is the `enzyme` control's value: `none` (no restriction step, so no
* enzymes and no refusal), a preset's name, a kit's, or `custom`, which reads
* `sites` — a list separated by commas or spaces, each item a preset's name or
* alias, a kit's, or a site with one `^`. The same enzyme named twice is cut
* once, so `DpnII, MboI` is DpnII's fragments rather than each cut counted
* twice.
*/
export function enzymesFor(choice: string, sites: string): Enzyme[] | { refusal: string } {
const value = choice.trim()
if (value === '' || value === 'none') return []
if (value !== 'custom') {
const kit = kitNamed(value)
if (kit !== undefined) return KITS[kit].map((name) => enzymeByName(name) as Enzyme)
const enzyme = enzymeByName(value)
return enzyme === undefined
? { refusal: `“${value}” is not an enzyme or a kit this knows. Pick one, or type its site under custom.` }
: [enzyme]
}
const items = sites.split(/[\s,;]+/).filter((item) => item !== '')
if (items.length === 0) return { refusal: 'Type an enzyme or a site, such as DpnII or G^ANTC.' }
const out: Enzyme[] = []
const seen = new Set<string>()
const keep = (enzyme: Enzyme): void => {
const key = `${enzyme.site}^${enzyme.cut}`
if (seen.has(key)) return
seen.add(key)
out.push(enzyme)
}
for (const item of items) {
const kit = kitNamed(item)
if (kit !== undefined) {
for (const name of KITS[kit]) keep(enzymeByName(name) as Enzyme)
continue
}
const named = enzymeByName(item)
if (named !== undefined) {
keep(named)
continue
}
const site = parseSite(item)
if ('refusal' in site) return site
keep(site)
}
if (out.length > MOST_ENZYMES) {
return { refusal: `That is ${out.length} enzymes, and one digest takes ${MOST_ENZYMES} at most.` }
}
return out
}
/** What a set of enzymes is called: `DpnII`, `DpnII+HinfI`, `G^ANTC`. */
export function enzymeLabel(enzymes: readonly Enzyme[]): string {
return enzymes.map((enzyme) => enzyme.name).join('+')
}
/** The longest site of a set, less one: what a piece of a reference must overlap the next by. */
export function siteOverlap(enzymes: readonly Enzyme[]): number {
let longest = 0
for (const enzyme of enzymes) longest = Math.max(longest, enzyme.site.length)
return Math.max(0, longest - 1)
}
// ---------------------------------------------------------------------------
// Cuts
// ---------------------------------------------------------------------------
/**
* The cuts of one piece or one sequence, three parallel arrays in the order of
* the top-strand cut.
*
* `top` is ascending and unique. Where two enzymes of one digest cut the top
* strand at one position, the first of them in the list is the one kept — on
* every real pair of sites that happens on, their bottom cuts agree too.
*/
export interface CutSites {
/** The top-strand cut `c` of each site: the first base after it, 0-based. */
top: Int32Array
/** The bottom-strand cut `c'` of the same site, in top-strand coordinates. */
bottom: Int32Array
/**
* Which enzyme of the digest made each cut, as a small index: its place in
* the list {@link cutsIn} was given, or, read back by {@link cutsOfResult},
* the order the enzymes' names first appear in.
*/
enzyme: Uint8Array
}
/** One enzyme compiled once, however many pieces of however many sequences it is run over. */
interface Compiled {
motif: Motif
/** Added to a match's start, on either strand, for the top-strand cut: `k`. */
offset: number
}
const COMPILED = new Map<string, Compiled>()
function compiledOf(enzyme: Enzyme): Compiled {
const key = `${enzyme.site}^${enzyme.cut}`
const held = COMPILED.get(key)
if (held !== undefined) return held
const motif = compileMotif(enzyme.site)
if (isMotifRefusal(motif)) throw new Error(`${enzyme.name}: ${motif.reason}`)
const made = { motif, offset: enzyme.cut }
COMPILED.set(key, made)
return made
}
/**
* A cut and the enzyme that made it, as one number that sorts by the cut: the
* cut times 256 plus the enzyme's position, exact in a double for any cut a
* 32-bit coordinate can hold. One typed array, sorted once, rather than three
* kept in step by hand.
*/
const ENZYME_SLOTS = 256
/**
* Sort `keys[0, n)`, keep the first of each top-strand cut, and lay them out
* as {@link CutSites}, dropping a cut outside `(lo, hi)`.
*/
function unpack(keys: Float64Array, n: number, shifts: Int32Array, lo: number, hi: number): CutSites {
const view = keys.subarray(0, n).sort()
let kept = 0
let last = Number.NaN
for (let i = 0; i < view.length; i++) {
const top = Math.floor(view[i] / ENZYME_SLOTS)
if (top === last || !(top > lo && top < hi)) continue
last = top
view[kept++] = view[i]
}
const out: CutSites = {
top: new Int32Array(kept),
bottom: new Int32Array(kept),
enzyme: new Uint8Array(kept),
}
for (let i = 0; i < kept; i++) {
const top = Math.floor(view[i] / ENZYME_SLOTS)
const enzyme = view[i] - top * ENZYME_SLOTS
out.top[i] = top
out.bottom[i] = top + shifts[enzyme]
out.enzyme[i] = enzyme
}
return out
}
/**
* Every cut the enzymes make in one piece of a sequence, in the order of the
* top-strand cut and unique by it, with `offset` added so the positions are the
* sequence's.
*
* Both strands are searched, and a palindrome — every preset — once
* (`findMotif`'s rule), so a site gives one cut. Several enzymes are merged: a
* position two of them cut is one boundary, which is what a fragment BED made
* from both digests says.
*
* Nothing is dropped for lying at an end, because a piece does not know where
* its sequence ends: {@link contigCuts} drops a cut at the sequence's first
* base or at its end. A caller reading a reference in pieces overlaps each with
* the next by {@link siteOverlap} and merges them with {@link contigCuts}, which
* finds every site exactly once whatever the seams.
*/
export function cutsIn(bases: Uint8Array, enzymes: readonly Enzyme[], offset = 0): CutSites {
let keys = new Float64Array(Math.max(16, bases.length >> 7))
let n = 0
enzymes.forEach((enzyme, at) => {
const compiled = compiledOf(enzyme)
const { hits } = findMotif(bases, compiled.motif, { bothStrands: true, offset })
if (n + hits.length > keys.length) {
const grown = new Float64Array(Math.max(keys.length * 2, n + hits.length))
grown.set(keys.subarray(0, n))
keys = grown
}
// `s + k` on either strand: a reverse match is cut where a forward one is
// (the module docstring), so the strand the site was read on is not asked.
for (const hit of hits) keys[n++] = (hit.start + compiled.offset) * ENZYME_SLOTS + at
})
const shifts = Int32Array.from(enzymes, bottomShift)
return unpack(keys, n, shifts, Number.NEGATIVE_INFINITY, Number.POSITIVE_INFINITY)
}
/**
* One sequence's cuts, from the pieces it was read in: merged, de-duplicated,
* and without a cut whose top strand is cut at `0` or at `length` or past it,
* which would be a fragment of no bases.
*
* A site lying wholly inside the overlap of two pieces is found by both, and
* its cut can fall on either side of the seam, so the pieces are merged by
* sorting rather than by appending.
*/
export function contigCuts(pieces: readonly CutSites[], length: number): CutSites {
let total = 0
for (const piece of pieces) total += piece.top.length
const keys = new Float64Array(total)
const shifts = new Int32Array(ENZYME_SLOTS)
let n = 0
for (const piece of pieces) {
for (let i = 0; i < piece.top.length; i++) {
keys[n++] = piece.top[i] * ENZYME_SLOTS + piece.enzyme[i]
shifts[piece.enzyme[i]] = piece.bottom[i] - piece.top[i]
}
}
return unpack(keys, n, shifts, 0, length)
}
/**
* `[0, …cuts, length]`: one strand's fragment boundaries, from its cuts in the
* order of the top-strand cut — {@link CutSites.top} for `B`,
* {@link CutSites.bottom} for `B'`.
*
* Nothing is dropped, so the two arrays a sequence's cuts give are the same
* length and fragment `i` is one fragment in both. Each boundary is kept at or
* after the one before and inside `[0, length]`: the top cuts already are, and
* so are the bottom cuts of one enzyme or of a kit, but two typed sites whose
* overhangs run in opposite directions can put one bottom cut before its
* neighbour's, and a fragment cannot end before it starts.
*/
export function boundaries(cuts: Int32Array, length: number): Int32Array {
const out = new Int32Array(cuts.length + 2)
let last = 0
for (let i = 0; i < cuts.length; i++) {
last = Math.min(length, Math.max(last, cuts[i]))
out[i + 1] = last
}
out[cuts.length + 1] = length
return out
}
/**
* The fragment a 0-based position lies in: `searchsorted(bounds, x, 'right') - 1`,
* clamped to `[0, fragments - 1]`, for the boundaries of the strand the end
* was read on — `B` for a `+` end, `B'` for a `-` one.
*
* The clamp is `pairtools restrict`'s (`find_rfrag`), checked rather than
* assumed: inside the sequence the two rules are one, and outside it pairtools
* puts a position before the start in the first fragment and one at the end or
* past it in the last, where an unclamped search would name a fragment that
* does not exist.
*/
export function fragmentOf(bounds: Int32Array, x: number): number {
// The last boundary at or below x, by halving a window that starts at it:
// a step is taken or not by a comparison rather than a branch, which on a
// `.pairs` whose ends land anywhere is most of what the search costs.
let lo = 0
let n = bounds.length
while (n > 1) {
const half = n >> 1
lo = bounds[lo + half] <= x ? lo + half : lo
n -= half
}
return Math.min(lo, Math.max(0, bounds.length - 2))
}
// ---------------------------------------------------------------------------
// Pairs
// ---------------------------------------------------------------------------
/** What a pair is, read against the fragments its ends lie in. */
export type PairClass = 'trans' | 'dangling' | 'self-circle' | 'same-strand' | 're-ligation' | 'valid'
/** Every class, in the order a table and a legend list them. */
export const PAIR_CLASSES: readonly PairClass[] = [
'valid',
're-ligation',
'dangling',
'self-circle',
'same-strand',
'trans',
]
/** The strand pair of two ends, as a table's column says it. */
export type Orientation = '++' | '+-' | '-+' | '--'
export const ORIENTATIONS: readonly Orientation[] = ['++', '+-', '-+', '--']
/**
* The class of a pair, from the fragments its two ends lie in — each placed
* by its own strand's cut ({@link fragmentOf}) — and their strands; `a` is the
* end with the smaller position, and `cis` false makes it `trans` whatever the
* rest says.
*
* - The same fragment, read towards each other (`+ -`): **dangling**, an end
* that was never cut or never ligated.
* - The same fragment, read away from each other (`- +`): **self-circle**, a
* fragment ligated to itself.
* - The same fragment on one strand (`+ +`, `- -`): **same-strand**.
* - Neighbouring fragments read towards each other: **re-ligation**, two
* neighbours joined back where they were cut.
* - Anything else: **valid**.
*/
export function classify(
cis: boolean,
fragA: number,
strandA: string,
fragB: number,
strandB: string,
): PairClass {
if (!cis) return 'trans'
if (fragA === fragB) {
if (strandA === strandB) return 'same-strand'
return strandA === '+' ? 'dangling' : 'self-circle'
}
return Math.abs(fragB - fragA) === 1 && strandA === '+' && strandB === '-' ? 're-ligation' : 'valid'
}
/** The classes' positions in {@link PAIR_CLASSES}, which is what the counter keys its counts on. */
const VALID = 0
const RELIGATION = 1
const DANGLING = 2
const SELF_CIRCLE = 3
const SAME_STRAND = 4
const TRANS = 5
/**
* {@link classify} on strand codes — `0` is `+`, `1` is `-` — for a cis pair,
* as a position in {@link PAIR_CLASSES}: the same rules, with no string made.
*/
function classCode(fragA: number, strandA: number, fragB: number, strandB: number): number {
if (fragA === fragB) {
if (strandA === strandB) return SAME_STRAND
return strandA === 0 ? DANGLING : SELF_CIRCLE
}
return Math.abs(fragB - fragA) === 1 && strandA === 0 && strandB === 1 ? RELIGATION : VALID
}
/** Everything a {@link RestrictionCounter} counts, as plain data that crosses a worker. */
export interface RestrictionStats {
/** {@link enzymeLabel} of the enzymes the cuts came from. */
enzymes: string
/** Pairs classified: both ends mapped, on sequences of the reference, not duplicates. */
counted: number
/** Pairs that would have counted but have an end on a sequence the reference has not got. */
unassigned: number
classes: Record<PairClass, number>
/**
* Each class by the strands of its two ends: `a` then `b` for a cis pair,
* ordered by position, and the file's order for a trans one.
*/
byOrientation: Record<PairClass, Record<Orientation, number>>
/** `distEdges()`: the left edge of every distance bin. */
distanceEdges: number[]
/** Ends per distance bin, by the end's strand. */
distance: { '+': number[]; '-': number[] }
fragments: {
/** Fragments on the reference's sequences. */
total: number
/** Of them, the ones at least one counted end lies in. */
seen: number
/** `0`, `1`, `2`, `3–4`, `5–8`, … through the last bin with a fragment in it. */
binLabels: string[]
/** Fragments per bin of {@link binLabels}: how many ends each fragment holds. */
counts: number[]
}
}
/**
* Distances below this are binned by a table rather than a search: most of a
* four-cutter's fragments are shorter, and it is 64 kB once per counter.
*/
const DIRECT_DISTANCES = 1 << 16
/** `0` for `+`, `1` for `-`, `-1` for anything else — which is an end not mapped. */
function strandCode(strand: string): number {
return strand === '+' ? 0 : strand === '-' ? 1 : -1
}
/** The bin of the ends-per-fragment histogram a count falls in: `0`, `1`, `2`, `3–4`, `5–8`, …. */
function endsBin(count: number): number {
return count === 0 ? 0 : 33 - Math.clz32(count - 1)
}
function endsBinLabel(bin: number): string {
if (bin <= 2) return String(bin)
const hi = 2 ** (bin - 1)
return `${hi / 2 + 1}–${hi}`
}
/**
* The restriction-fragment statistics of a `.pairs`, one record at a time.
*
* Built once from the cuts of the reference — {@link cutsOfResult} rebuilds them
* from a `restriction` run — and the lengths of its sequences; then
* {@link add} per record and {@link result} once. The sequences of `cuts` are
* the reference: each one with a length in `lengths` is cut into its fragments,
* one with no cuts is one fragment, and a pair with an end anywhere else is
* {@link RestrictionStats.unassigned}.
*
* **Names are matched as everywhere else (§5.9).** The cuts are in the
* reference's spelling and a `.pairs` writes its own, so a name that is not one
* of the reference's is looked up once through `contigIndexOf`'s ladder — `I`
* finds `chrI` — and the answer is kept, so a file of twenty million records
* pays for each distinct spelling once. `lengths` is read the same way, so it
* may be the `.pairs` header's or the map's.
*
* **Nothing is allocated per record.** The counts are typed arrays sized at
* construction, the strands are compared rather than concatenated, and the last
* name seen on each side is remembered, since a sorted `.pairs` repeats the
* same pair of chromosomes for millions of lines.
*/
export class RestrictionCounter {
private readonly names: string[] = []
/** `B` and `B'` per sequence. */
private readonly top: Int32Array[] = []
private readonly bottom: Int32Array[] = []
/** Where each sequence's fragments start in {@link ends}. */
private readonly firstFragment: number[] = []
private readonly index = new Map<string, number>()
private readonly ends: Uint32Array
private readonly classCounts = new Float64Array(PAIR_CLASSES.length * ORIENTATIONS.length)
private readonly edges = distEdges()
private readonly plus: Float64Array
private readonly minus: Float64Array
private counted = 0
private unassigned = 0
private last1 = ''
private lastAt1 = -1
private last2 = ''
private lastAt2 = -1
/**
* The sequence and the fragment the last end on each side fell in. A sorted
* `.pairs` walks the first end along a chromosome, so the next one is
* usually in the same fragment — on either strand, since fragment `i` is one
* fragment in both — and asking that first saves the search.
*/
private nearContig1 = -1
private nearAt1 = 0
private nearContig2 = -1
private nearAt2 = 0
/** The distance bin of every distance below {@link DIRECT_DISTANCES}, looked up rather than searched. */
private readonly distanceBin: Uint8Array
constructor(
cuts: ReadonlyMap<string, CutSites>,
lengths: ReadonlyMap<string, number>,
private readonly label = '',
) {
const lengthNames = [...lengths.keys()]
let fragments = 0
for (const [name, sites] of cuts) {
let length = lengths.get(name)
if (length === undefined) {
const at = contigIndexOf(lengthNames, name)
length = at < 0 ? undefined : lengths.get(lengthNames[at])
}
// A sequence whose length nobody gave cannot be cut into fragments — its
// last one has no end — so it is not part of the reference this counts
// against, and its pairs are unassigned rather than guessed at.
if (length === undefined || !(length > 0) || this.index.has(name)) continue
const top = boundaries(sites.top, length)
this.index.set(name, this.names.length)
this.names.push(name)
this.top.push(top)
this.bottom.push(boundaries(sites.bottom, length))
this.firstFragment.push(fragments)
fragments += top.length - 1
}
this.ends = new Uint32Array(fragments)
this.plus = new Float64Array(this.edges.length)
this.minus = new Float64Array(this.edges.length)
this.distanceBin = new Uint8Array(DIRECT_DISTANCES)
for (let d = 0, bin = 0; d < DIRECT_DISTANCES; d++) {
while (bin + 1 < this.edges.length && this.edges[bin + 1] <= d) bin++
this.distanceBin[d] = bin
}
}
/** A sequence's index among the reference's, or `-1`; resolved once per spelling. */
private indexOf(name: string): number {
const held = this.index.get(name)
if (held !== undefined) return held
const at = contigIndexOf(this.names, name)
this.index.set(name, at)
return at
}
/**
* One record: both ends as the file writes them, `pos` 1-based, and whether
* the record is a duplicate (`pair_type` `DD`).
*
* An end on `!`, or with a strand other than `+` or `-`, is unmapped and the
* pair is not counted; nor is a duplicate. A mapped pair with an end on a
* sequence the reference has not got is {@link RestrictionStats.unassigned}.
*/
add(
chrom1: string,
pos1: number,
strand1: string,
chrom2: string,
pos2: number,
strand2: string,
duplicate: boolean,
): void {
if (duplicate) return
const s1 = strandCode(strand1)
const s2 = strandCode(strand2)
if (s1 < 0 || s2 < 0 || chrom1 === '!' || chrom2 === '!') return
// A position that is not a number is a record nobody can place.
if (!(pos1 === pos1) || !(pos2 === pos2)) return
let c1: number
if (chrom1 === this.last1) c1 = this.lastAt1
else {
c1 = this.indexOf(chrom1)
this.last1 = chrom1
this.lastAt1 = c1
}
let c2: number
if (chrom2 === this.last2) c2 = this.lastAt2
else {
c2 = this.indexOf(chrom2)
this.last2 = chrom2
this.lastAt2 = c2
}
if (c1 < 0 || c2 < 0) {
this.unassigned++
return
}
this.counted++
const x1 = pos1 - 1
const x2 = pos2 - 1
// Each end by its own strand's cut: a `-` read starts on the bottom strand.
// The fragment the last end on this side fell in is tried first: if `x` is
// inside it, it is the last boundary at or below `x`, which is the answer.
const b1 = s1 === 0 ? this.top[c1] : this.bottom[c1]
let f1 = this.nearAt1
if (c1 !== this.nearContig1 || !(x1 >= b1[f1] && x1 < b1[f1 + 1])) {
f1 = fragmentOf(b1, x1)
this.nearContig1 = c1
this.nearAt1 = f1
}
const b2 = s2 === 0 ? this.top[c2] : this.bottom[c2]
let f2 = this.nearAt2
if (c2 !== this.nearContig2 || !(x2 >= b2[f2] && x2 < b2[f2 + 1])) {
f2 = fragmentOf(b2, x2)
this.nearContig2 = c2
this.nearAt2 = f2
}
this.ends[this.firstFragment[c1] + f1]++
this.ends[this.firstFragment[c2] + f2]++
this.distanceOf(c1, f1, x1, s1)
this.distanceOf(c2, f2, x2, s2)
let code: number
let orientation: number
if (c1 !== c2) {
code = TRANS
orientation = s1 * 2 + s2
} else if (x2 < x1) {
code = classCode(f2, s2, f1, s1)
orientation = s2 * 2 + s1
} else {
code = classCode(f1, s1, f2, s2)
orientation = s1 * 2 + s2
}
this.classCounts[code * ORIENTATIONS.length + orientation]++
}
/**
* One end's distance to the next cut in the direction it reads, on its own
* strand's boundaries: `B[i+1] - x` on `+`, `x - B'[i] + 1` on `-`, so `1` is
* an end whose 5′ base is the last before the cut, and the two reads of a
* dangling end both give the fragment's length. An end placed outside its
* sequence lands in the first bin rather than nowhere.
*/
private distanceOf(contig: number, fragment: number, x: number, strand: number): void {
const d = strand === 0 ? this.top[contig][fragment + 1] - x : x - this.bottom[contig][fragment] + 1
let lo = 0
if (d >= 0 && d < DIRECT_DISTANCES) lo = this.distanceBin[d]
else if (d > 0) {
// `searchsorted(edges, d, 'right') - 1`, for a fragment longer than the table.
const edges = this.edges
let hi = edges.length
while (hi - lo > 1) {
const mid = (lo + hi) >> 1
if (edges[mid] <= d) lo = mid
else hi = mid
}
}
if (strand === 0) this.plus[lo]++
else this.minus[lo]++
}
/** Everything counted, as plain data. */
result(): RestrictionStats {
const classes = {} as Record<PairClass, number>
const byOrientation = {} as Record<PairClass, Record<Orientation, number>>
PAIR_CLASSES.forEach((name, code) => {
const row = {} as Record<Orientation, number>
let sum = 0
ORIENTATIONS.forEach((orientation, at) => {
const count = this.classCounts[code * ORIENTATIONS.length + at]
row[orientation] = count
sum += count
})
classes[name] = sum
byOrientation[name] = row
})
const histogram: number[] = []
let seen = 0
for (let f = 0; f < this.ends.length; f++) {
const count = this.ends[f]
if (count > 0) seen++
const bin = endsBin(count)
while (histogram.length <= bin) histogram.push(0)
histogram[bin]++
}
return {
enzymes: this.label,
counted: this.counted,
unassigned: this.unassigned,
classes,
byOrientation,
distanceEdges: [...this.edges],
distance: { '+': Array.from(this.plus), '-': Array.from(this.minus) },
fragments: {
total: this.ends.length,
seen,
binLabels: histogram.map((_, bin) => endsBinLabel(bin)),
counts: histogram,
},
}
}
}
// ---------------------------------------------------------------------------
// A `restriction` run, read back
// ---------------------------------------------------------------------------
/**
* The cuts a `restriction` task result holds, by sequence and on both strands,
* or `null` for a result with no sites output.
*
* Read from its intervals output, whose every interval is one cut: `start` is
* the top-strand cut — the first base after it — and the interval is that one
* base; `name` is the enzyme that made it, a preset's name or a typed site
* with its `^`, which is what gives the bottom-strand cut back
* ({@link bottomShift}). A kit's cuts are two enzymes' with two shifts, so the
* name is per cut rather than per run. Every sequence the run cut is listed in
* the set's `contigs`, a sequence with no cut among them, so a sequence missing
* from the map is one the run did not read rather than one with no sites.
*/
export function cutsOfResult(result: TaskResult): Map<string, CutSites> | null {
const output = result.outputs.find(
(one): one is IntervalsOutput =>
one.kind === 'intervals' && one.id.startsWith('restriction-') && one.payload.kind === 'features',
)
if (output === undefined || output.payload.kind !== 'features') return null
const set = output.payload.data
const out = new Map<string, CutSites>()
// One shift per distinct name: a run names its cuts with a handful of strings.
const shifts = new Map<string, number>()
const slots = new Map<string, number>()
for (let c = 0; c < set.contigs.length; c++) {
const from = set.contigAt[c] ?? 0
const to = set.contigAt[c + 1] ?? from
const top = set.start.slice(from, to)
const bottom = new Int32Array(top.length)
const enzyme = new Uint8Array(top.length)
for (let i = 0; i < top.length; i++) {
const name = set.name[from + i] ?? ''
let shift = shifts.get(name)
if (shift === undefined) {
const known = enzymeNamed(name)
// A name this build cannot read is placed by its top cut on both
// strands, which is the reading every blunt cutter has anyway.
shift = known === undefined ? 0 : bottomShift(known)
shifts.set(name, shift)
slots.set(name, Math.min(ENZYME_SLOTS - 1, slots.size))
}
bottom[i] = top[i] + shift
enzyme[i] = slots.get(name) ?? 0
}
out.set(set.contigs[c], { top, bottom, enzyme })
}
return out
}
|