@datagrok/bio 2.13.6 → 2.13.8

package/package.json

@@ -5,7 +5,7 @@
     "name": "Aleksandr Tanas",
     "email": "atanas@datagrok.ai"
   },
-  "version": "2.13.6",
+  "version": "2.13.8",
   "description": "Bioinformatics support (import/export of sequences, conversion, visualization, analysis). [See more](https://github.com/datagrok-ai/public/blob/master/packages/Bio/README.md) for details.",
   "repository": {
     "type": "git",
@@ -39,10 +39,10 @@
     "@biowasm/aioli": "^3.1.0",
     "@datagrok-libraries/bio": "^5.42.6",
     "@datagrok-libraries/chem-meta": "^1.2.5",
-    "@datagrok-libraries/math": "^1.1.5",
-    "@datagrok-libraries/ml": "^6.6.12",
-    "@datagrok-libraries/tutorials": "^1.3.12",
-    "@datagrok-libraries/utils": "^4.2.12",
+    "@datagrok-libraries/math": "^1.1.12",
+    "@datagrok-libraries/ml": "^6.6.17",
+    "@datagrok-libraries/tutorials": "^1.3.13",
+    "@datagrok-libraries/utils": "^4.2.22",
     "@webgpu/types": "^0.1.40",
     "ajv": "^8.12.0",
     "ajv-errors": "^3.0.0",

package/src/demo/bio01-similarity-diversity.ts

@@ -15,9 +15,8 @@ export async function demoBio01UI() {
   let df: DG.DataFrame;
 
   try {
-    const demoScript = new DemoScript(
-      'Similarity, Diversity',
-      'Sequence similarity tracking and evaluation dataset diversity');
+    const demoScript = new DemoScript('Similarity, Diversity',
+      'Sequence similarity tracking and evaluation dataset diversity', false, {autoStartFirstStep: true});
     await demoScript
       .step(`Load DNA sequences`, async () => {
         grok.shell.windows.showContextPanel = false;

package/src/demo/bio01a-hierarchical-clustering-and-sequence-space.ts

@@ -23,9 +23,9 @@ export async function demoBio01aUI() {
   const activityColName = 'activity';
 
   try {
-    const demoScript = new DemoScript(
-      'Sequence Space',
-      'Exploring sequence space of Macromolecules, comparison with hierarchical clustering results');
+    const demoScript = new DemoScript('Sequence Space',
+      'Exploring sequence space of Macromolecules, comparison with hierarchical clustering results',
+      false, {autoStartFirstStep: true});
     await demoScript
       .step(`Load DNA sequences`, async () => {
         [df, treeHelper, dendrogramSvc] = await Promise.all([

package/src/demo/bio01b-hierarchical-clustering-and-activity-cliffs.ts

@@ -27,9 +27,8 @@ export async function demoBio01bUI() {
   const dimRedMethod: DimReductionMethods = DimReductionMethods.UMAP;
 
   try {
-    const demoScript = new DemoScript(
-      'Activity Cliffs',
-      'Activity Cliffs analysis on Macromolecules data');
+    const demoScript = new DemoScript('Activity Cliffs', 'Activity Cliffs analysis on Macromolecules data', false,
+      {autoStartFirstStep: true});
     await demoScript
       .step(`Load DNA sequences`, async () => {
         grok.shell.windows.showContextPanel = false;

package/src/demo/bio03-atomic-level.ts

@@ -15,10 +15,7 @@ export async function demoBio03UI(): Promise<void> {
   let dlg: DG.Dialog;
 
   try {
-    await new DemoScript(
-      'Atomic Level',
-      'Atomic level structure of Macromolecules',
-    )
+    await new DemoScript('Atomic Level', 'Atomic level structure of Macromolecules', false, {autoStartFirstStep: true})
       .step(`Loading Macromolecules notation 'Helm'`, async () => {
         grok.shell.windows.showContextPanel = false;
         grok.shell.windows.showProperties = false;

package/src/demo/bio05-helm-msa-sequence-space.ts

@@ -38,9 +38,8 @@ export async function demoBio05UI(): Promise<void> {
   let pepseaDcStartPromise: Promise<void>;
 
   try {
-    const demoScript = new DemoScript(
-      'Helm, MSA, Sequence Space',
-      'MSA and composition analysis on Helm data');
+    const demoScript = new DemoScript('Helm, MSA, Sequence Space', 'MSA and composition analysis on Helm data',
+      false, {autoStartFirstStep: true});
     await demoScript
       .step(`Load peptides with non-natural aminoacids in 'HELM' notation`, async () => {
         [pepseaDcStatus, df] = await Promise.all([

package/src/package-test.ts

@@ -44,8 +44,9 @@ export {tests};
 //input: string category {optional: true}
 //input: string test {optional: true}
 //input: object testContext {optional: true}
+//input: bool stressTest {optional: true}
 //output: dataframe result
-export async function test(category: string, test: string, testContext: TestContext): Promise<DG.DataFrame> {
-  const data = await runTests({category, test, testContext});
+export async function test(category: string, test: string, testContext: TestContext, stressTest?: boolean): Promise<DG.DataFrame> {
+  const data = await runTests({category, test, testContext, stressTest});
   return DG.DataFrame.fromObjects(data)!;
 }

package/src/package.ts

@@ -78,6 +78,7 @@ import {CyclizedNotationProvider} from './utils/cyclized';
 import {getMolColumnFromHelm} from './utils/helm-to-molfile/utils';
 import {PackageSettingsEditorWidget} from './widgets/package-settings-editor-widget';
 import {getUserLibSettings, setUserLibSettings} from '@datagrok-libraries/bio/src/monomer-works/lib-settings';
+import { calculateScoresWithEmptyValues } from './utils/calculate-scores';
 
 export const _package = new BioPackage();
 
@@ -868,7 +869,7 @@ export function SubsequenceSearchTopMenu(macromolecules: DG.Column): void {
 export async function sequenceIdentityScoring(
   table: DG.DataFrame, macromolecule: DG.Column, reference: string
 ): Promise<DG.Column<number>> {
-  const scores = calculateScores(table, macromolecule, reference, SCORE.IDENTITY);
+  const scores = calculateScoresWithEmptyValues(table, macromolecule, reference, SCORE.IDENTITY);
   return scores;
 }
 
@@ -882,7 +883,7 @@ export async function sequenceIdentityScoring(
 export async function sequenceSimilarityScoring(
   table: DG.DataFrame, macromolecule: DG.Column, reference: string
 ): Promise<DG.Column<number>> {
-  const scores = calculateScores(table, macromolecule, reference, SCORE.SIMILARITY);
+  const scores = calculateScoresWithEmptyValues(table, macromolecule, reference, SCORE.SIMILARITY);
   return scores;
 }
 

package/src/tests/mm-distance-tests.ts

@@ -82,7 +82,7 @@ ATCGAATCGA`;
 
   test('needleman-blosum62-del', async () => {
     const df = mmDistanceFunctions[MmDistanceFunctionsNames.NEEDLEMANN_WUNSCH]({gapOpen: 8, gapExtend: 2});
-    _testDistance(prot3, prot4, df, -1.8);
+    _testDistance(prot3, prot4, df, -3.667);
   });
 
   test('needleman-custom-sub', async () => {
@@ -96,21 +96,21 @@ ATCGAATCGA`;
     const df = mmDistanceFunctions[MmDistanceFunctionsNames.NEEDLEMANN_WUNSCH](
       {scoringMatrix, alphabetIndexes, gapOpen: 1, gapExtend: 1},
     );
-    _testDistance(prot3, prot4, df, 0.8);
+    _testDistance(prot3, prot4, df, 0.667);
   });
 
   test('needleman-custom-zero-extend', async () => {
     const df = mmDistanceFunctions[MmDistanceFunctionsNames.NEEDLEMANN_WUNSCH](
       {scoringMatrix, alphabetIndexes, gapOpen: 1, gapExtend: 0},
     );
-    _testDistance(prot5, prot6, df, 0.714);
+    _testDistance(prot5, prot6, df, 0.333);
   });
 
   test('needleman-custom-half-extend', async () => {
     const df = mmDistanceFunctions[MmDistanceFunctionsNames.NEEDLEMANN_WUNSCH](
       {scoringMatrix, alphabetIndexes, gapOpen: 2, gapExtend: 1},
     );
-    _testDistance(prot5, prot6, df, 1.286);
+    _testDistance(prot5, prot6, df, 1.667);
   });
 
   test('needleman-custom-same-extend', async () => {
@@ -121,7 +121,8 @@ ATCGAATCGA`;
       const seq1 = Array(10000).fill('FWRY').join('');
       const seq2 = Array(10000).fill('FYWRRY').join('');
       _testDistance(seq1, seq2, df, 0.667);
-    } else { _testDistance(prot5, prot6, df, 1.143); }
+    } else
+      _testDistance(prot5, prot6, df, 1.333);
   }, {benchmark: true});
 });
 

package/src/tests/pepsea-tests.ts

@@ -40,7 +40,7 @@ category('PepSeA', () => {
     const tgtMsaCol = df.getCol('MSA');
     for (let i = 0; i < resMsaCol!.length; ++i)
       expect(resMsaCol!.get(i) == tgtMsaCol.get(i), true);
-  }, {timeout: 60000 /* docker */});
+  }, {timeout: 60000 /* docker */, stressTest: true});
 
   test('stderr', async () => {
     const logger = new TestLogger();
@@ -50,7 +50,7 @@ category('PepSeA', () => {
     const tgtMsaCol = df.getCol('MSA');
     expectArray(resMsaCol!.toList(), tgtMsaCol.toList());
     expect(logger.warningList[0].message, pepseaStderrWarningList);
-  }, {timeout: 60000 /* docker */});
+  }, {timeout: 60000 /* docker */, stressTest: true});
 
   test('error', async () => {
     const logger = new TestLogger();

package/src/utils/calculate-scores.ts

@@ -0,0 +1,13 @@
+import * as DG from 'datagrok-api/dg';
+import { calculateScores, SCORE } from '@datagrok-libraries/bio/src/utils/macromolecule/scoring';
+
+export async function calculateScoresWithEmptyValues(
+    table: DG.DataFrame, macromolecule: DG.Column, reference: string, scoring: SCORE
+): Promise<DG.Column<number>> {
+    const scores = await calculateScores(table, macromolecule, reference, scoring);
+    for (let i = 0; i < scores.length; i++) {
+        if (macromolecule.isNone(i))
+            scores.set(i, null, false);
+    }
+    return scores;
+}

package/src/utils/monomer-cell-renderer.ts

@@ -105,6 +105,9 @@ export class MonomerCellRenderer extends DG.GridCellRenderer {
     const color = palette.get(s);
 
     g.fillStyle = color;
-    g.fillText(monomerToShort(s, 6), x + (w / 2), y + (h / 2), w);
+    //cell width of monomer should dictate how many characters can be displayed
+    // for width 40, 6 characters can be displayed (0.15 is 6 / 40)
+    const maxChars = Math.max(2, Math.floor(w * 0.15));
+    g.fillText(monomerToShort(s, maxChars), x + (w / 2), y + (h / 2), w);
   }
 }

package/dist/317.js

@@ -1,2 +0,0 @@
-var bio;(()=>{"use strict";var e,n,t={9260:(e,n,t)=>{var r;!function(e){e.EUCLIDEAN="EUCLIDEAN",e.MANHATTAN="MANHATTAN"}(r||(r={}));const a={[r.EUCLIDEAN]:function(e){return`\n        var sum = 0.0;\n        for (var i = 0u; i < ${e}; i = i + 1u) {\n            sum = sum + distances[i] * distances[i] * computeInfo.weights[i] * computeInfo.weights[i];\n        }\n        return sqrt(sum);\n    `},[r.MANHATTAN]:function(e){return`\n        var sum = 0.0;\n        for (var i = 0u; i < ${e}; i = i + 1u) {\n            sum = sum + abs(distances[i]) * computeInfo.weights[i];\n        }\n        return sum;\n    `}};var i;!function(e){e.HAMMING="Hamming",e.EUCLIDEAN="Euclidean",e.MANHATTAN="Manhattan",e.TANIMOTO="Tanimoto",e.LEVENSTEIN="Levenshtein",e.NEEDLEMAN_WUNSCH="Needlemann-Wunsch",e.MONOMER_CHEMICAL_DISTANCE="Monomer chemical distance",e.SOKAL="Sokal",e.COSINE="Cosine",e.ASYMMETRIC="Asymmetric",e.Difference="Difference",e.OneHot="One-Hot"}(i||(i={}));const o={[i.HAMMING]:function(e,n){return`\n  let aLength: u32 = computeInfo.entrySizes[${n}][aIndex];\n  let bLength: u32 = computeInfo.entrySizes[${n}][bIndex];\n  let maxLength: u32 = max(aLength, bLength);\n  let minLength: u32 = min(aLength, bLength);\n  let sizeDiff: u32 = maxLength - minLength;\n  \n  let maxIntDistance = ceil(maxDistance * f32(maxLength)) - f32(sizeDiff);\n\n  var diff: f32 = 0.0;\n  for (var i = 0u; i < ${e}; i = i + 1u) {\n      diff = diff + f32(a[i] != b[i]);\n      if (diff > maxIntDistance) {\n            return 1.0;\n        }\n  }\n  diff += f32(sizeDiff);\n  return diff / ${e};\n  `},[i.EUCLIDEAN]:function(e,n){return`\n  var dist: f32 = 0.0;\n  for (var i = 0u; i < ${e}; i = i + 1u) {\n    dist = dist + f32(a[i] - b[i]) * f32(a[i] - b[i]);\n  }\n  return sqrt(dist);\n  `},[i.MANHATTAN]:function(e,n){return`\n  var dist: f32 = 0.0;\n  for (var i = 0u; i < ${e}; i = i + 1u) {\n    dist = dist + abs(f32(a[i] - b[i]));\n  }\n  return dist;\n  `},[i.TANIMOTO]:function(e,n){return`\n  var onBitsa: u32 = 0u;\n  var onBitsb: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      onBitsa = onBitsa + countOneBits(a[i]);\n      onBitsb = onBitsb + countOneBits(b[i]);\n  }\n\n  if (onBitsa == 0u && onBitsb == 0u) {\n      return 0.0;\n  }\n\n  let totalOnBits = onBitsa + onBitsb;\n  var commonBits: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      commonBits = commonBits + countOneBits(a[i] & b[i]);\n  }\n\n  return 1.0 - f32(commonBits) / f32(totalOnBits - commonBits);\n  `},[i.LEVENSTEIN]:function(e,n){return`\n  let aLength: u32 = computeInfo.entrySizes[${n}][aIndex];\n  let bLength: u32 = computeInfo.entrySizes[${n}][bIndex];\n  let maxLength: u32 = max(aLength, bLength);\n  let minLength: u32 = min(aLength, bLength);\n\n  let maxIntDistance = ceil(maxDistance * f32(maxLength));\n\n  // we will store two arrays as matrix and swap the working indices per pass.\n  // this way we can reduce memory usage per computation to just O(aLength)\n  // the grid will have aLength + 1 columns and bLength + 1 rows\n  // this will be guaranteed by iteration, but the array sizes must be known at compile time, so we will use a fixed size of maxArraySize\n  var dynamicPassMat: array<array<f32, ${e+1}u>, 2>; // initialize to 0\n  \n  var prevIndex: u32 = 0;\n  var curIndex: u32 = 1; // we will swap these indices per pass\n\n  // initialize the first row\n  for (var i = 0u; i <= aLength; i = i + 1u) {\n      dynamicPassMat[prevIndex][i] = f32(i);\n  }\n\n  // iterate over the rows\n  for (var i = 1u; i <= bLength; i = i + 1u) {\n      dynamicPassMat[curIndex][0] = f32(i);\n      var minEntry: f32 = f32(maxLength);\n      let prevRow = &dynamicPassMat[prevIndex];\n      let curRow = &dynamicPassMat[curIndex];\n      let bMon = u32(b[i - 1]);\n      for (var j = 1u; j <= aLength; j = j + 1u) {\n          var cost: f32 = f32(a[j - 1] != bMon);\n          var res: f32 = min(\n              min(\n                (*prevRow)[j] + 1.0, // deletion\n                (*curRow)[j - 1] + 1.0, // insertion\n              ),\n              (*prevRow)[j - 1] + cost // substitution\n              );\n              (*curRow)[j] = res;\n          if (res < minEntry) {\n              minEntry = res;\n          }\n      }\n      // swap the indices\n      let temp: u32 = prevIndex;\n      prevIndex = curIndex;\n      curIndex = temp;\n      if (minEntry > maxIntDistance) {\n          return 1.0;\n      }\n  }\n\n  return dynamicPassMat[prevIndex][aLength] / f32(maxLength);\n  `},[i.NEEDLEMAN_WUNSCH]:function(e,n){return`\n      let aLength: u32 = computeInfo.entrySizes[${n}][aIndex];\n      let bLength: u32 = computeInfo.entrySizes[${n}][bIndex];\n      let maxLength: u32 = max(aLength, bLength);\n      let minLength: u32 = min(aLength, bLength);\n  \n      let maxIntDistance = ceil(maxDistance * f32(maxLength));\n      // we will store two arrays as matrix and swap the working indices per pass.\n      // this way we can reduce memory usage per computation to just O(aLength)\n      // the grid will have aLength + 1 columns and bLength + 1 rows\n      // this will be guaranteed by iteration, but the array sizes must be known at compile time, so we will use a fixed size of maxArraySize\n      var dynamicPassMat: array<array<f32, ${e+1}u>, 2>; // initialize to 0\n      \n      // we need to keep track of which operation led to the current cell\n      // i.e. whether we came from the left, top or diagonal to assign gap open/gap extend penalty\n      var verticalGaps: array<u32, ${e+1}u>;\n      var horizontalGaps: array<u32, ${e+1}u>;\n\n      let gapOpenPenalty: f32 = suppInfo.gapOpenPenalty${n};\n      let gapExtensionPenalty: f32 = suppInfo.gapExtensionPenalty${n};\n      var prevIndex: u32 = 0;\n      var curIndex: u32 = 1; // we will swap these indices per pass\n      // initialize the first row\n      for (var i = 0u; i <= aLength; i = i + 1u) {\n          dynamicPassMat[prevIndex][i] = gapOpenPenalty + f32(i - 1) * gapExtensionPenalty;\n          dynamicPassMat[curIndex][i] = 0.0;\n      }\n      dynamicPassMat[0][0] = 0.0;\n\n      let simMatrix = &suppInfo.similarityMatrix${n}; // using pointers make things faster\n      // iterate over the rows\n      for (var i = 1u; i <= bLength; i = i + 1u) {\n          let prevRow = &dynamicPassMat[prevIndex];\n          let curRow = &dynamicPassMat[curIndex];\n          (*curRow)[0] = gapOpenPenalty + f32(i - 1) * gapExtensionPenalty;\n          var minEntry: f32 = f32(maxLength);\n          let monB = u32(b[i - 1]);\n          for (var j = 1u; j <= aLength; j = j + 1u) {\n              let monA = u32(a[j - 1]);\n              \n              let cost: f32 = (*prevRow)[j - 1] + 1f - (*simMatrix)[monA][monB];\n              var top = (*prevRow)[j]; // deletion\n              if (verticalGaps[j] > 0) {\n                  top = top + gapExtensionPenalty;\n              } else {\n                  top = top + gapOpenPenalty;\n              }\n              var left = (*curRow)[j - 1]; // insertion\n              if (horizontalGaps[j - 1] > 0) {\n                  left = left + gapExtensionPenalty;\n              } else {\n                  left = left + gapOpenPenalty;\n              }\n              var res: f32 = min(\n                  min(\n                      top, // deletion\n                      left, // insertion\n                  ),\n                  cost // substitution\n                  );\n                  (*curRow)[j] = res;\n              if (res < minEntry) {\n                  minEntry = res;\n              }\n              // update the horizontal and vertical gaps\n              if (res == cost) {\n                  verticalGaps[j] = 0;\n                  horizontalGaps[j] = 0;\n              } else if (res == left) {\n                  verticalGaps[j] = 0;\n                  horizontalGaps[j] = 1;\n              } else {\n                  verticalGaps[j] = 1;\n                  horizontalGaps[j] = 0;\n              }\n          }\n          // swap the indices\n          let temp: u32 = prevIndex;\n          prevIndex = curIndex;\n          curIndex = temp;\n          if (minEntry > maxIntDistance) {\n              return 1.0;\n          }\n      }\n      return dynamicPassMat[prevIndex][aLength] / f32(maxLength);\n  \n  `},[i.MONOMER_CHEMICAL_DISTANCE]:function(e,n){return`\n  let aLength: u32 = computeInfo.entrySizes[${n}][aIndex];\n  let bLength: u32 = computeInfo.entrySizes[${n}][bIndex];\n  let maxLength: u32 = max(aLength, bLength);\n  let minLength: u32 = min(aLength, bLength);\n  let sizeDiff: u32 = maxLength - minLength;\n  \n  let maxIntDistance = ceil(maxDistance * f32(maxLength)) - f32(sizeDiff);\n\n  let simMatrix = &(suppInfo.similarityMatrix${n}); // using pointers make things faster\n  var diff: f32 = 0.0;\n  for (var i = 0u; i < ${e}; i = i + 1u) {\n      diff = diff + 1.0 - (*simMatrix)[u32(a[i])][u32(b[i])];\n      if (diff > maxIntDistance) {\n            return 1.0;\n        }\n  }\n  diff += f32(sizeDiff);\n  return diff / ${e};\n  `},[i.SOKAL]:function(e,n){return`\n  var onBitsa: u32 = 0u;\n  var onBitsb: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      onBitsa = onBitsa + countOneBits(a[i]);\n      onBitsb = onBitsb + countOneBits(b[i]);\n  }\n  let total = onBitsa + onBitsb;\n  if (total == 0u) {\n    return 1.0;\n  }\n  var commonBits: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      commonBits = commonBits + countOneBits(a[i] & b[i]);\n  }\n  return 1.0 - f32(commonBits) / f32(total * 2 - commonBits * 3);\n  `},[i.COSINE]:function(e,n){return`\n  var onBitsa: u32 = 0u;\n  var onBitsb: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      onBitsa = onBitsa + countOneBits(a[i]);\n      onBitsb = onBitsb + countOneBits(b[i]);\n  }\n  let total = onBitsa * onBitsb; // p.s. here total is taken by multiplying\n  if (total == 0u) {\n    return 1.0;\n  }\n  var commonBits: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      commonBits = commonBits + countOneBits(a[i] & b[i]);\n  }\n  return 1.0 - f32(commonBits) / sqrt(f32(total));\n  `},[i.ASYMMETRIC]:function(e,n){return`\n  var onBitsa: u32 = 0u;\n  var onBitsb: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      onBitsa = onBitsa + countOneBits(a[i]);\n      onBitsb = onBitsb + countOneBits(b[i]);\n  }\n  let min = min(onBitsa, onBitsb);\n  if (min == 0u) {\n    return 1.0;\n  }\n  var commonBits: u32 = 0u;\n  for (var i = 0u; i < ${e}u; i = i + 1u) {\n      commonBits = commonBits + countOneBits(a[i] & b[i]);\n  }\n  return 1.0 - f32(commonBits) / f32(min);\n  `},[i.Difference]:function(e,n){return`\n    let range = suppInfo.range${n};\n    return f32(abs(f32(a[0]) - f32(b[0])) / range);\n  `},[i.OneHot]:function(e,n){return`\n  let aLength: u32 = computeInfo.entrySizes[${n}][aIndex];\n  let bLength: u32 = computeInfo.entrySizes[${n}][bIndex];\n  if (aLength != bLength) {\n    return 1.0;\n  }\n  for (var i = 0u; i < aLength; i = i + 1u) {\n    if(a[i] != b[i]) {\n      return 1.0;\n    }\n  }\n  return 0.0;\n  `}},s={[i.HAMMING]:e=>Math.ceil(e/30),[i.EUCLIDEAN]:e=>Math.ceil(e/30),[i.MANHATTAN]:e=>Math.ceil(e/30),[i.TANIMOTO]:e=>Math.ceil(e/60),[i.SOKAL]:e=>Math.ceil(e/60),[i.COSINE]:e=>Math.ceil(e/60),[i.ASYMMETRIC]:e=>Math.ceil(e/60),[i.LEVENSTEIN]:e=>Math.ceil(e*e/60),[i.NEEDLEMAN_WUNSCH]:e=>Math.ceil(e*e/60),[i.MONOMER_CHEMICAL_DISTANCE]:e=>Math.ceil(e/25),[i.Difference]:e=>1,[i.OneHot]:e=>Math.ceil(e/40)},u={STRING:new Set([i.HAMMING,i.LEVENSTEIN,i.NEEDLEMAN_WUNSCH,i.MONOMER_CHEMICAL_DISTANCE,i.OneHot]),UINT32ARRAY:new Set([i.HAMMING,i.EUCLIDEAN,i.MANHATTAN,i.MONOMER_CHEMICAL_DISTANCE,i.LEVENSTEIN,i.NEEDLEMAN_WUNSCH,i.TANIMOTO,i.COSINE,i.SOKAL,i.ASYMMETRIC,i.OneHot,i.Difference]),INT32ARRAY:new Set([i.EUCLIDEAN,i.MANHATTAN,i.OneHot,i.Difference]),FLOAT32ARRAY:new Set([i.EUCLIDEAN,i.MANHATTAN,i.Difference]),NUMBER:new Set([i.EUCLIDEAN,i.MANHATTAN,i.Difference]),BITARRAY:new Set([i.TANIMOTO,i.COSINE,i.SOKAL,i.ASYMMETRIC])};var l=function(e,n,t,r){return new(t||(t=Promise))((function(a,i){function o(e){try{u(r.next(e))}catch(e){i(e)}}function s(e){try{u(r.throw(e))}catch(e){i(e)}}function u(e){var n;e.done?a(e.value):(n=e.value,n instanceof t?n:new t((function(e){e(n)}))).then(o,s)}u((r=r.apply(e,n||[])).next())}))};let f=null,c=null;function d(){return l(this,void 0,void 0,(function*(){if(!f&&(f=yield navigator.gpu.requestAdapter({powerPreference:"high-performance"}),null==f))return null;let e=!1;if(c&&(c.lost.then((()=>{e=!0})),yield new Promise((e=>setTimeout(e,10)))),!c||e){const e=1e9,n=f.limits,t=n.maxBufferSize,r=n.maxStorageBufferBindingSize;try{return c=yield f.requestDevice({requiredLimits:{maxBufferSize:Math.min(t,e),maxStorageBufferBindingSize:Math.min(r,e)}}),c}catch(e){return console.error("Failed to create device with required limits",e),c=yield f.requestDevice(),c}}return c}))}function p(e,n=.8,t,a,o,l){return f=this,c=void 0,h=function*(){const f=yield d();if(!f)return null;const c=Object.values(i);if(t.some((e=>!c.includes(e))))throw new Error("Invalid distance metrics provided: "+t.join(", "));if(!Object.values(r).includes(a))throw new Error("Invalid aggregation function provided: "+a);const p=1-n;if(l.length!==e.length||l.length!==t.length||l.length!==o.length)throw new Error("Options, weigths and distance functions must be provided for each column");if(e.some((n=>n.length!==e[0].length)))throw new Error("All entry lists must be the same length");const h=e.length,g=e[0].length,y=e.map(((e,n)=>function(e,n=i.HAMMING,t,r={gapOpenPenalty:1,gapExtensionPenalty:.6}){var a,o;let l=null;const f=e.some((e=>"string"==typeof e))?(l="STRING",e.map((e=>new Uint32Array(e.split("").map((e=>e.charCodeAt(0))))))):e.some((e=>"number"==typeof e))?(l="NUMBER",e.map((e=>new Float32Array([e])))):"object"==typeof e[0]&&e.some((e=>"_data"in e&&"_length"in e))?(l="BITARRAY",e.map((e=>e._data))):e.some((e=>e instanceof Float32Array))?(l="FLOAT32ARRAY",e):e.some((e=>e instanceof Uint32Array))?(l="UINT32ARRAY",e):e.some((e=>e instanceof Int32Array))?(l="INT32ARRAY",e):void 0;if(!f||!l)throw new Error("Invalid entry type, could not determine entry type from input list");const c=f[0]instanceof Int32Array?"INT32ARRAY":f[0]instanceof Float32Array?"FLOAT32ARRAY":"UINT32ARRAY",d=new Uint32Array(f.map((e=>e.length)));if(!u[l]||!u[l].has(n))throw new Error(`Distance metric '${n}' not supported for entry type '${l}'`);const p=d.reduce(((e,n)=>Math.max(e,n)),0),m=s[n](p),h="INT32ARRAY"===c?Int32Array:"FLOAT32ARRAY"===c?Float32Array:Uint32Array,g=new h(f.length*p);f.forEach(((e,n)=>{g.set(e,n*p)}));let y="",A=0,b="FLOAT32ARRAY",w=null;if(n===i.NEEDLEMAN_WUNSCH||n===i.MONOMER_CHEMICAL_DISTANCE){let e=r.scoringMatrix&&r.alphabetIndexes?Object.keys(r.alphabetIndexes).reduce(((e,n)=>Math.max(e,n.charCodeAt(0))),0):-1;if(!r.alphabetIndexes||!r.scoringMatrix){for(let n=0;n<g.length;n++)g[n]>e&&(e=g[n]);r.scoringMatrix=new Array(e+1).fill(null).map((()=>new Array(e+1).fill(0))),r.alphabetIndexes={};for(let e=0;e<r.scoringMatrix.length;e++)r.scoringMatrix[e][e]=1,r.alphabetIndexes[String.fromCharCode(e)]=e}const n=(e+1)*(e+1),i=new Array(e+1).fill(null).map((()=>new Float32Array(e+1)));for(let n=0;n<e+1;n++)i[n][n]=1;const s=r.alphabetIndexes;for(const e of Object.keys(s))for(const n of Object.keys(s))e!==n&&(i[e.charCodeAt(0)][n.charCodeAt(0)]=r.scoringMatrix[s[e]][s[n]]);A=2+n,b="FLOAT32ARRAY",w=new Float32Array(A),w[0]=null!==(a=r.gapOpenPenalty)&&void 0!==a?a:1,w[1]=null!==(o=r.gapExtensionPenalty)&&void 0!==o?o:.6;let u=2;for(let e=0;e<i.length;e++)w.set(i[e],u),u+=i[e].length;y=`\n            gapOpenPenalty${t}: f32,\n            gapExtensionPenalty${t}: f32,\n            similarityMatrix${t}: array<array<f32, ${e+1}>, ${e+1}>`}else if(n===i.Difference){if(!r.range||"number"!=typeof r.range||r.range<=0){const e=g.reduce(((e,n)=>Math.min(e,n)),g[0]),n=g.reduce(((e,n)=>Math.max(e,n)),g[0]);r.range=n-e}r.range<=0&&(r.range=1),A=1,b="FLOAT32ARRAY",w=new Float32Array([r.range]),y=`\n            range${t}: f32`}const x=g instanceof Int32Array?"i32":g instanceof Float32Array?"f32":"u32",E=`data${t}: array<array<${x}, ${p}>, ${f.length}>`;return{flatSourceArray:g,sourceArraySize:g.length,maxEntryLen:p,arraySizes:d,complexity:m,suppInfoBuffer:w,suppInfoSize:A,suppInfoType:b,suppInfoStructWgsl:y,entryType:l,dataTypeWGSL:x,dataStructWgsl:E,EncodedArrayConstructor:h}}(e,t[n],n,l[n])));if(0===h)throw new Error("No columns provided. Please provide at least one column of data.");1===h&&(a=r.MANHATTAN);let A=y.map((e=>e.suppInfoStructWgsl)).filter((e=>!!e&&""!=e)).join(",\n"),b=!1;A&&""!=A.trim()||(b=!0,A="\ndummy: f32\n");const w=y.map((e=>e.dataStructWgsl)).filter((e=>!!e&&""!=e)).join(",\n"),x=new Uint32Array(h*g);y.forEach(((e,n)=>{x.set(e.arraySizes,n*g)}));const E=1e4,I=100,M=y.reduce(((e,n)=>e+n.complexity),0),N=Math.ceil(1e4/M),v=Math.ceil(Math.sqrt(Math.ceil(100))),B=10*v,S=g*(g-1)/2,U=Math.ceil(S/E),C=f.createShaderModule({label:"Sparse matrix compute shader",code:`\n    // each thread will perform 100 iterations at one time, comparing 100 pairs of entries.\n    // in total, each thread will perform at most ${U} comparisons.\n    // first is the result struct, containing is, js, and distances. each array with length of 100,\n    // and also integer for how many pairs were found to be below threshold.\n    struct SparseResult {\n      i: array<array<u32, 100>, 10000>,\n      j: array<array<u32, 100>, 10000>,\n      distances: array<array<f32, 100>, 10000>,\n      found: array<u32, 10000>,\n      done: array<u32, 10000>\n    }\n    // struct for the data\n    struct ComputeInfo {\n      // start at cols and rows, and end at cols and rows for each thread, these will be calculated on cpu and passed to gpu.\n      startAtCols: array<u32, 10000>,\n      startAtRows: array<u32, 10000>,\n      endAtCols: array<u32, 10000>,\n      endAtRows: array<u32, 10000>,\n\n      // the ACTUALLY sizes of each entry\n      entrySizes: array<array<u32, ${g}>, ${h}>,\n      // the weights for each entry\n      weights: array<f32, ${h}>,\n      // the data for each entry\n      ${w} // an example of the dataWgsl would be:\n      //data0: array<array<u32,20>,100>,\n      //data1: array<array<u32,20>,100>\n    }\n\n    // struct for the supplementary information\n    struct SuppInfo {\n      // struct containing all the supplementary info, like scoring matrix, alphabet indexes, range, etc.\n      ${A}\n    };\n\n    @group(0) @binding(0) var<storage, read_write> computeInfo: ComputeInfo;\n    @group(0) @binding(1) var<storage, read_write> suppInfo: SuppInfo;\n    @group(0) @binding(2) var<storage, read_write> results: SparseResult;\n    @compute @workgroup_size(10, 10) fn calcSparseMatrix(\n      @builtin(global_invocation_id) id: vec3<u32>\n    ) {\n      ${b?"let otherDummy = suppInfo.dummy * 2;":""} // just to make sure that the suppInfo is not optimized out\n      let threadCol = id.x;\n      let threadRow = id.y;\n      let linearIndex = threadRow * ${B} + threadCol;\n      if (linearIndex >= 10000) {\n        return; // if we are out of bounds, return\n      }      \n      var startAtCol: u32 = computeInfo.startAtCols[linearIndex];\n      var startAtRow: u32 = computeInfo.startAtRows[linearIndex];\n      let endAtCol: u32 = min(computeInfo.endAtCols[linearIndex], ${g}u);\n      let endAtRow: u32 = min(computeInfo.endAtRows[linearIndex], ${g}u);\n      let is = &results.i[linearIndex];\n      let js = &results.j[linearIndex];\n      let distances = &results.distances[linearIndex];\n      results.found[linearIndex] = 0; // initialize the found counter\n      var found: u32 = 0;\n      if (results.done[linearIndex] > 0) {\n        return; // if we are done, return\n      }\n      for (var i = 0; i < ${N}; i++) {\n        if (startAtCol >= endAtCol && startAtRow >= endAtRow) {\n          results.done[linearIndex] = 1;\n          break;\n        }\n        if (found >= 100) {\n          break;\n        }\n        let dist = combinedDistance(startAtCol, startAtRow);\n        if (dist <= ${p}) {\n          (*is)[found] = startAtCol;\n          (*js)[found] = startAtRow;\n          (*distances)[found] = dist;\n          found = found + 1;\n        }\n        startAtCol = startAtCol + 1;\n        if (startAtCol >= ${g}u) {\n          startAtRow += 1;\n          startAtCol = startAtRow + 1;\n        }\n      }\n      results.found[linearIndex] = found;\n      // update the startAtCols and startAtRows\n      computeInfo.startAtCols[linearIndex] = startAtCol;\n      computeInfo.startAtRows[linearIndex] = startAtRow;\n\n    }\n\n    // this will generate the distance script for each distance metric and then combine them into one\n    ${m(t,y.map((e=>e.maxEntryLen)),p,a)}\n\n\n    `}),O=f.createComputePipeline({label:"sparse matrix compute pipeline",layout:"auto",compute:{module:C,entryPoint:"calcSparseMatrix"}}),L=new Uint32Array(E),P=new Uint32Array(E),R=new Uint32Array(E),T=new Uint32Array(E),_=Math.floor(S/E);let j=0,$=1;console.time("GPUthreadStarts");for(let e=0;e<E;e++){const n=9999===e?S-1:(e+1)*_,t=g-2-Math.floor(Math.sqrt(-8*n+4*g*(g-1)-7)/2-.5),r=n-g*t+Math.floor((t+1)*(t+2)/2);L[e]=$,P[e]=j,R[e]=r,T[e]=t,j=t,$=r}console.timeEnd("GPUthreadStarts");const D=4e4+g*h+h+y.reduce(((e,n)=>e+n.sourceArraySize),0),G=y.reduce(((e,n)=>e+n.suppInfoSize),0),z=1e6,k=D*Uint32Array.BYTES_PER_ELEMENT;let Y=k;const F=15&k;0!==F&&(Y+=16-F);const H=f.createBuffer({label:"compute info buffer",size:Y,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0}),K=H.getMappedRange();let W=0;new Uint32Array(K,W,E).set(L),W+=E*Uint32Array.BYTES_PER_ELEMENT,new Uint32Array(K,W,E).set(P),W+=E*Uint32Array.BYTES_PER_ELEMENT,new Uint32Array(K,W,E).set(R),W+=E*Uint32Array.BYTES_PER_ELEMENT,new Uint32Array(K,W,E).set(T),W+=E*Uint32Array.BYTES_PER_ELEMENT,new Uint32Array(K,W,x.length).set(x),W+=x.length*Uint32Array.BYTES_PER_ELEMENT,new Float32Array(K,W,h).set(o),W+=h*Float32Array.BYTES_PER_ELEMENT;for(const e of y){const n=e.EncodedArrayConstructor,t=e.sourceArraySize;new n(K,W,t).set(e.flatSourceArray),W+=t*n.BYTES_PER_ELEMENT}H.unmap();const q=G*Uint32Array.BYTES_PER_ELEMENT;let X=q;const V=15&q;0!==V&&(X+=16-V),X=Math.max(X,16);const Q=f.createBuffer({label:"supp info buffer",size:X,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0}),J=Q.getMappedRange();let Z=0;for(const e of y)e.suppInfoBuffer&&e.suppInfoBuffer.byteLength>0&&e.suppInfoSize>0&&(new("UINT32ARRAY"===e.suppInfoType?Uint32Array:Float32Array)(J,Z,e.suppInfoBuffer.length).set(e.suppInfoBuffer),Z+=e.suppInfoBuffer.byteLength);0===Z&&new Uint32Array(J,0,4).set([1,1,1,1]),Q.unmap();const ee=302e4*Uint32Array.BYTES_PER_ELEMENT;let ne=ee;const te=15&ee;0!==te&&(ne+=16-te);const re=f.createBuffer({label:"results buffer",size:ne,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC}),ae=f.createBindGroup({label:"bindGroup for sparse matrix buffer",layout:O.getBindGroupLayout(0),entries:[{binding:0,resource:{buffer:H}},{binding:1,resource:{buffer:Q}},{binding:2,resource:{buffer:re}}]}),ie=f.createBuffer({label:"results out buffer",size:re.size,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST}),oe=[],se=[],ue=[];let le=!1;for(;!le;){const e=f.createCommandEncoder({label:"distance encoder"}),n=e.beginComputePass({label:"distance compute pass"});n.setPipeline(O),n.setBindGroup(0,ae),n.dispatchWorkgroups(v,v),n.end(),e.copyBufferToBuffer(re,0,ie,0,ie.size);const t=e.finish();f.queue.submit([t]),yield f.queue.onSubmittedWorkDone(),yield ie.mapAsync(GPUMapMode.READ);const r=ie.getMappedRange();let a=0;const i=new Uint32Array(r,a,z);a+=z*Uint32Array.BYTES_PER_ELEMENT;const o=new Uint32Array(r,a,z);a+=z*Uint32Array.BYTES_PER_ELEMENT;const s=new Float32Array(r,a,z);a+=z*Float32Array.BYTES_PER_ELEMENT;const u=new Uint32Array(r,a,E);a+=E*Uint32Array.BYTES_PER_ELEMENT,le=new Uint32Array(r,a,E).every((e=>1===e));const l=u.reduce(((e,n)=>e+n),0),c=new Uint32Array(l),d=new Uint32Array(l),p=new Float32Array(l);let m=0;for(let e=0;e<u.length;e++){const n=u[e];0!==n&&(c.set(i.subarray(e*I,e*I+n),m),d.set(o.subarray(e*I,e*I+n),m),p.set(s.subarray(e*I,e*I+n),m),m+=n)}oe.push(c),se.push(d),ue.push(p),ie.unmap()}const fe=oe.reduce(((e,n)=>e+n.length),0),ce=new Uint32Array(fe),de=new Uint32Array(fe),pe=new Float32Array(fe);let me=0;for(let e=0;e<oe.length;e++)ce.set(oe[e],me),de.set(se[e],me),pe.set(ue[e],me),me+=oe[e].length;return H.destroy(),Q.destroy(),re.destroy(),ie.destroy(),{i:ce,j:de,distance:pe}},new((p=void 0)||(p=Promise))((function(e,n){function t(e){try{a(h.next(e))}catch(e){n(e)}}function r(e){try{a(h.throw(e))}catch(e){n(e)}}function a(n){var a;n.done?e(n.value):(a=n.value,a instanceof p?a:new p((function(e){e(a)}))).then(t,r)}a((h=h.apply(f,c||[])).next())}));var f,c,p,h}function m(e,n,t,r){return e.map(((e,r)=>`\n        fn distanceScript${r}(aIndex: u32, bIndex: u32) -> f32 {\n          let a = computeInfo.data${r}[aIndex];\n          let b = computeInfo.data${r}[bIndex];\n          let maxDistance: f32 = ${t};\n          ${o[e](n[r],r)}\n        }\n      `)).join("\n")+"\n"+`\n      fn combinedDistance(aIndex: u32, bIndex: u32) -> f32 {\n        var distances: array<f32, ${e.length}>;\n        ${e.map(((e,n)=>`distances[${n}] = distanceScript${n}(aIndex, bIndex);`)).join("\n")}\n        ${a[r](e.length)}\n      }\n    \n    `}var h=t(5731);async function g(e,n,t){const r=e.length,a=[],i=[];for(let o=0;o<r;o++)for(let s=o+1;s<r;s++){const r=e[o],u=e[s];n[r]?.[u]>=t&&(a.push(o),i.push(s))}return function(e,n,t){const r=new Float32Array(t.length).fill(0).map((()=>10*Math.random())),a=new Float32Array(t.length).fill(0).map((()=>10*Math.random())),i=new Float32Array(t.length).fill(0),o=new Float32Array(t.length).fill(0);for(let s=0;s<100;s++){const u=1-s/100;i.fill(0),o.fill(0);for(let t=0;t<e.length;t++){const s=e[t],l=n[t],f=r[s]-r[l],c=a[s]-a[l];Math.abs(f)>=1&&(i[s]-=u*f,i[l]+=u*f),Math.abs(c)>=1&&(o[s]-=u*c,o[l]+=u*c)}for(let e=0;e<t.length;e++){const n=Math.sqrt(i[e]*i[e]+o[e]*o[e]);n>0&&(r[e]+=i[e]/n*u,a[e]+=o[e]/n*u)}}let s=r[0],u=a[0],l=r[0],f=a[0];for(let e=1;e<t.length;e++)s=Math.min(s,r[e]),u=Math.min(u,a[e]),l=Math.max(l,r[e]),f=Math.max(f,a[e]);let c=l-s,d=f-u;0===c&&(c=l);for(let e=0;e<t.length;e++)r[e]=(r[e]-s)/c/2+.5;0===d&&(d=f);for(let e=0;e<t.length;e++)a[e]=(a[e]-u)/d/2+.5;return{embedX:r,embedY:a}}(a,i,e)}function y(e,n,t,r){return a=this,i=void 0,s=function*(){const a=r,i=Math.ceil(a/100),o=Math.ceil(Math.sqrt(i)),s=10*o,u=e.createShaderModule({label:"colwise-normalize",code:`\n        @group(0) @binding(0) var<storage, read_write> knnSimilarities : array<f32>;\n        @group(0) @binding(1) var<storage, read> offsets : array<u32>;\n        @compute @workgroup_size(10, 10) fn normalize(\n            @builtin(global_invocation_id) id: vec3<u32>,\n          ) {\n            let row = id.x;\n            let col = id.y;\n            let index = row * ${s} + col;\n            if (index >= ${r}) {\n                return;\n            }\n            let offsetBegin = offsets[index];\n            let offsetEnd = offsets[index + 1];\n            var sum = 0.0;\n            if (offsetEnd - offsetBegin == 0) {\n                return;\n            }\n            for (var i = offsetBegin; i < offsetEnd; i = i + 1) {\n                sum = sum + knnSimilarities[i];\n            }\n            if (sum > 0.0) {\n              for (var i = offsetBegin; i < offsetEnd; i = i + 1) {\n                  knnSimilarities[i] = knnSimilarities[i] / sum;\n              }\n            }\n          }\n        \n        `}),l=e.createComputePipeline({label:"hamming compute pipeline",layout:"auto",compute:{module:u,entryPoint:"normalize"}}),f=e.createBuffer({label:"simmilarities info buffer",size:n.byteLength,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0});new Float32Array(f.getMappedRange()).set(n),f.unmap();const c=e.createBuffer({label:"offsets info buffer",size:t.byteLength,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0});new Uint32Array(c.getMappedRange()).set(t),c.unmap();const d=e.createBindGroup({layout:l.getBindGroupLayout(0),entries:[{binding:0,resource:{buffer:f}},{binding:1,resource:{buffer:c}}]}),p=e.createCommandEncoder(),m=p.beginComputePass();m.setPipeline(l),m.setBindGroup(0,d),m.dispatchWorkgroups(o,o),m.end();const h=e.createBuffer({label:"out similarity buffer",size:n.byteLength,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST});p.copyBufferToBuffer(f,0,h,0,n.byteLength),e.queue.submit([p.finish()]),yield e.queue.onSubmittedWorkDone(),yield h.mapAsync(GPUMapMode.READ);const g=new Float32Array(h.getMappedRange());n.set(g),h.unmap(),f.destroy(),c.destroy(),h.destroy()},new((o=void 0)||(o=Promise))((function(e,n){function t(e){try{u(s.next(e))}catch(e){n(e)}}function r(e){try{u(s.throw(e))}catch(e){n(e)}}function u(n){var a;n.done?e(n.value):(a=n.value,a instanceof o?a:new o((function(e){e(a)}))).then(t,r)}u((s=s.apply(a,i||[])).next())}));var a,i,o,s}function A(e,n=2){for(let t=0;t<e.length;t++)e[t]=Math.pow(e[t],n)}function b(e,n,t,r,a){return i=this,o=void 0,u=function*(){const i=9e4,o=Math.ceil(900),s=Math.ceil(Math.sqrt(o)),u=10*s,l=Math.floor(Math.max(Math.log10(a),2))+1,f=Math.pow(10,-l),c=new Float32Array(n.length),d=e.createShaderModule({label:"expand",code:`\n        struct SparseKNN {\n            knnSimilarities: array<f32, ${n.length}>,\n            knnIndexes: array<u32, ${t.length}>,\n            offsets: array<u32, ${r.length}>,\n            rowIndexes: array<u32, ${t.length}>,\n        }\n\n        @group(0) @binding(0) var<storage, read_write> sparseKNN: SparseKNN;\n        @group(0) @binding(1) var<storage, read_write> resultSimBlock: array<f32, 90000>;\n        @group(0) @binding(2) var<storage, read_write> startAt: u32;\n        @compute @workgroup_size(10, 10) fn expand(\n            @builtin(global_invocation_id) id: vec3<u32>,\n          ) {\n            let col = id.x;\n            let row = id.y;\n            let index = row * ${u} + col;\n            if (index >= 90000) {\n                return;\n            }\n            let workingIndex = index + startAt;\n            if (workingIndex >= ${t.length}) {\n                return;\n            }\n            \n            let rowIdx = sparseKNN.rowIndexes[workingIndex];\n            let colIdx = sparseKNN.knnIndexes[workingIndex];\n            let offsetBeginRow = sparseKNN.offsets[rowIdx];\n            let offsetEndRow = sparseKNN.offsets[rowIdx + 1];\n            let offsetBeginCol = sparseKNN.offsets[colIdx];\n            let offsetEndCol = sparseKNN.offsets[colIdx + 1];\n            var sum = 0.0;\n            for (var i = offsetBeginRow; i < offsetEndRow; i = i + 1) {\n                for(var j = offsetBeginCol; j < offsetEndCol; j = j + 1) {\n                    if (sparseKNN.knnIndexes[i] == sparseKNN.knnIndexes[j]) {\n                        sum = sum + sparseKNN.knnSimilarities[i] * sparseKNN.knnSimilarities[j];\n                        break;\n                    }\n                }\n            }\n            if (sum > ${f}) {\n                resultSimBlock[index] = sum;\n            } else {\n                resultSimBlock[index] = 0.0;\n            }\n        }\n    `}),p=e.createComputePipeline({label:"expand compute pipeline",layout:"auto",compute:{module:d,entryPoint:"expand"}});let m=4*(n.length+t.length+r.length+t.length);const h=15&m;0!==h&&(m+=16-h);const g=e.createBuffer({label:"sparse knn buffer",size:m,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0}),y=g.getMappedRange();new Float32Array(y,0,n.length).set(n),new Uint32Array(y,4*n.length,t.length).set(t),new Uint32Array(y,4*(n.length+t.length),r.length).set(r);const A=function(e){const n=new Uint32Array(e[e.length-1]);for(let t=0;t<e.length-1;t++)for(let r=e[t];r<e[t+1];r++)n[r]=t;return n}(r);new Uint32Array(y,4*(n.length+t.length+r.length),A.length).set(A),g.unmap();const b=e.createBuffer({label:"start end buffer",size:4,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST,mappedAtCreation:!0});new Uint32Array(b.getMappedRange()).set([0]),b.unmap();const w=e.createBuffer({label:"result block buffer",size:36e4,usage:GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST}),x=e.createBindGroup({layout:p.getBindGroupLayout(0),entries:[{binding:0,resource:{buffer:g}},{binding:1,resource:{buffer:w}},{binding:2,resource:{buffer:b}}]}),E=e.createBuffer({label:"out block buffer",size:w.size,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST});for(let n=0;n<Math.ceil(t.length/i);n++){const r=n*i,a=Math.min(i,t.length-r);e.queue.writeBuffer(b,0,new Uint32Array([r]));const o=e.createCommandEncoder({label:"expand encoder"}),u=o.beginComputePass({label:"expand compute pass"});u.setPipeline(p),u.setBindGroup(0,x),u.dispatchWorkgroups(s,s),u.end(),o.copyBufferToBuffer(w,0,E,0,E.size),e.queue.submit([o.finish()]),yield e.queue.onSubmittedWorkDone(),yield E.mapAsync(GPUMapMode.READ);const l=new Float32Array(E.getMappedRange(),0,a);c.set(l,r),E.unmap()}return g.destroy(),b.destroy(),w.destroy(),E.destroy(),{KNNIndexes:t,KNNSimilarities:c,indexOffsets:r}},new((s=void 0)||(s=Promise))((function(e,n){function t(e){try{a(u.next(e))}catch(e){n(e)}}function r(e){try{a(u.throw(e))}catch(e){n(e)}}function a(n){var a;n.done?e(n.value):(a=n.value,a instanceof s?a:new s((function(e){e(a)}))).then(t,r)}a((u=u.apply(i,o||[])).next())}));var i,o,s,u}function w(e,n){const t=function(e,n){const t=new Uint32Array(n);for(let n=0;n<e.i.length;n++)t[e.i[n]]++,t[e.j[n]]++;for(let e=0;e<n;e++)t[e]++;return t}(e,n),r=function(e){const n=new Uint32Array(e.length+1);let t=0;for(let r=0;r<e.length;r++)n[r]=t,t+=e[r];return n[e.length]=t,n}(t),a=new Uint32Array(n).fill(1),i=new Uint32Array(2*e.i.length+n),o=new Float32Array(2*e.i.length+n);for(let e=0;e<n;e++)i[r[e]]=e,o[r[e]]=1;for(let n=0;n<e.i.length;n++){const t=e.i[n],s=e.j[n],u=1-e.distance[n];i[r[t]+a[t]]=s,o[r[t]+a[t]]=u,a[t]++,i[r[s]+a[s]]=t,o[r[s]+a[s]]=u,a[s]++}return{KNNIndexes:i,KNNSimilarities:o,indexOffsets:r}}Object.prototype.toString;const x={expandFactor:2,maxIterations:5,inflateFactor:2,multFactor:1};class E{constructor(e={}){this._options={...x,...e}}async transform(e,n){let t=this.toObjectForm(e);if(this._options.maxIterations>0){this.addLoops(t,n),this.normalize(t);for(let e=0;e<this._options.maxIterations;e++)t=this.expand(t,n),this.inflate(t),this.normalize(t)}const{clusters:r,is:a,js:i}=this.assignClusters(t,n);this.correctClusters(r);const o=await this.layout(r,t,n);return{clusters:r,embedX:o.embedX,embedY:o.embedY,is:a,js:i}}async transformWebGPU(e,n){if(0===this._options.maxIterations)return this.transform(e,n);const t=await function(e,n,t=5,r=2){return a=this,i=void 0,s=function*(){const a=yield d();if(!a)throw new Error("no gpu device found");const i=w(e,n);yield y(a,i.KNNSimilarities,i.indexOffsets,n);let o=i;for(let e=0;e<t;e++){const e=yield b(a,o.KNNSimilarities,o.KNNIndexes,o.indexOffsets,n);A(e.KNNSimilarities,r),yield y(a,e.KNNSimilarities,e.indexOffsets,n),o=e}return o},new((o=void 0)||(o=Promise))((function(e,n){function t(e){try{u(s.next(e))}catch(e){n(e)}}function r(e){try{u(s.throw(e))}catch(e){n(e)}}function u(n){var a;n.done?e(n.value):(a=n.value,a instanceof o?a:new o((function(e){e(a)}))).then(t,r)}u((s=s.apply(a,i||[])).next())}));var a,i,o,s}(e,n,this._options.maxIterations,this._options.inflateFactor),r=this.csrToSparseObject(t,n),{clusters:a,is:i,js:o}=this.assignClusters(r,n);this.correctClusters(a);const s=await this.layout(a,r,n);return{clusters:a,embedX:s.embedX,embedY:s.embedY,is:i,js:o}}correctClusters(e){const n={};for(const t of e)n[t]||(n[t]=0),n[t]++;const t=Object.keys(n).map(Number).sort(((e,t)=>n[t]-n[e])),r={};t.forEach(((e,n)=>r[e]=n+1));for(let n=0;n<e.length;n++)e[n]=r[e[n]]}csrToSparseObject(e,n){const t=Math.floor(Math.max(Math.log10(n),2))+1,r=1/Math.pow(10,t),a={};for(let t=0;t<n;t++){a[t]={};for(let n=e.indexOffsets[t];n<e.indexOffsets[t+1];n++){const i=e.KNNIndexes[n];i<=t||e.KNNSimilarities[n]<r||(a[t][i]=e.KNNSimilarities[n])}}return a}async layout(e,n,t){const r=new Float32Array(t).fill(0),a=new Float32Array(t).fill(0),i={};e.forEach(((e,n)=>{i[e]||(i[e]=[]),i[e].push(n)}));let o=0;const s=Object.keys(i);s.sort(((e,n)=>i[n].length-i[e].length));let u=6,l=0;for(const e of s){const t=i[e],s=await g(t,n,.001);o===Math.ceil(u/1.5)&&(o=0,l+=5/u,u=Math.ceil(1.5*u));const f=o%u*5/u*1.5;for(let e=0;e<s.embedX.length;e++)r[t[e]]=5*s.embedX[e]/u+f,a[t[e]]=5*s.embedY[e]/u+l;o++}return{embedX:r,embedY:a}}mergeClusters(e,n,t){const r=e[n],a=e[t];for(let n=0;n<e.length;n++)e[n]===a&&(e[n]=r)}assignClusters(e,n){let t=0;const r=[],a=[],i=Math.floor(Math.max(Math.log10(n),2))+1,o=Math.pow(10,i),s=new Array(n).fill(-1);for(const n of Object.keys(e))for(const i of Object.keys(e[n]))Math.round(e[n][i]*o)/o>0&&e[n][i]!==Number(n)&&Number(i)>Number(n)&&(r.push(Number(n)),a.push(Number(i)),-1!==s[Number(n)]&&-1!==s[Number(i)]?s[Number(n)]!==s[Number(i)]&&this.mergeClusters(s,Number(n),Number(i)):-1!==s[Number(n)]?s[Number(i)]=s[Number(n)]:-1!==s[Number(i)]?s[Number(n)]=s[Number(i)]:(t++,s[Number(n)]=t,s[Number(i)]=t));for(let e=0;e<s.length;e++)-1===s[e]&&(t++,s[e]=t);return{clusters:s,is:new Uint32Array(r),js:new Uint32Array(a)}}assignClustersCSR(e,n){let t=0;const r=[],a=[],i=Math.floor(Math.max(Math.log10(n),2))+1,o=1/Math.pow(10,i),s=new Array(n).fill(-1),u=new Uint32Array(n+1);let l=0;u[0]=0;for(let i=0;i<n;i++){for(let n=e.indexOffsets[i];n<e.indexOffsets[i+1];n++){const u=e.KNNIndexes[n];u<=i||e.KNNSimilarities[n]<=o||(r.push(i),a.push(u),l++,-1!==s[i]&&-1!==s[u]?s[i]!==s[u]&&this.mergeClusters(s,i,u):-1!==s[i]?s[u]=s[i]:-1!==s[u]?s[i]=s[u]:(t++,s[i]=t,s[u]=t))}u[i+1]=l}for(let e=0;e<s.length;e++)-1===s[e]&&(t++,s[e]=t);return{clusters:s,is:new Uint32Array(r),js:new Uint32Array(a),correctedOffsets:u}}toObjectForm(e){const n={};for(let t=0;t<e.i.length;t++)n[e.i[t]]||(n[e.i[t]]={}),n[e.i[t]][e.j[t]]=1-e.distance[t],n[e.j[t]]||(n[e.j[t]]={}),n[e.j[t]][e.i[t]]=1-e.distance[t];return n}addLoops(e,n){for(let t=0;t<n;t++)e[t]||(e[t]={}),e[t][t]=this._options.multFactor}normalize(e){for(const n of Object.keys(e)){const t=e[n];let r=0;for(const e of Object.keys(t))r+=t[e];if(0!==r)for(const a of Object.keys(t))e[n][a]/=r}}expand(e,n){const t={},r=Math.floor(Math.max(Math.log10(n),2))+1,a=Math.pow(10,r);for(let r=0;r<n;r++)if(e[r]){t[r]??(t[r]={});for(let i=r;i<n;i++){if(!e[r]?.[i])continue;const n=this.getExpandValue(e,r,i);Math.round(n*a)/a>0&&(t[r][i]=n,t[i]||(t[i]={}),t[i][r]=n)}}return t}inflate(e){for(const n of Object.keys(e)){const t=e[n];for(const r of Object.keys(t))e[n][r]=Math.pow(e[n][r],this._options.inflateFactor)}}getExpandValue(e,n,t){let r=0;const a=Object.keys(e[n]??{}),i=Object.keys(e[t]??{});for(const o of a)i.includes(o)&&(r+=e[n][o]*e[t][o]);return r}}onmessage=async e=>{const{data:n,threshold:t,weights:r,aggregationMethod:a,distanceFnArgs:i,distanceFns:o,maxIterations:s,useWebGPU:u,inflate:l}=e.data;console.time("sparse matrix");let f=null;if(u)try{f=await p(n,t/100,o,a,r,i)}catch(e){console.error(e)}f||(u&&console.error("WEBGPU sparse matrix calculation failed, falling back to CPU implementation"),f=await(new h.p).calcMultiColumn(n,o,t/100,i,r,a)),console.timeEnd("sparse matrix");const c=new E({maxIterations:s??5,inflateFactor:l??2});console.time("MCL");let d=null;if(u)try{d=await c.transformWebGPU(f,n[0].length)}catch(e){console.error("webGPU MCL failed, falling back to CPU implementation"),console.error(e)}d||(d=await c.transform(f,n[0].length)),console.timeEnd("MCL"),postMessage({res:d})}},606:(e,n,t)=>{t.d(n,{Qt:()=>a,gD:()=>r}),t(6066);const r=e=>null==e;function a(e,n,t,r){if(t>e[e.length-1])return;const a=e.findIndex((e=>t<e));e.pop(),e.splice(a,0,t),n.pop(),n.splice(a,0,r)}}},r={};function a(e){var n=r[e];if(void 0!==n)return n.exports;var i=r[e]={exports:{}};return t[e](i,i.exports,a),i.exports}a.m=t,a.x=()=>{var e=a.O(void 0,[731],(()=>a(9260)));return a.O(e)},e=[],a.O=(n,t,r,i)=>{if(!t){var o=1/0;for(f=0;f<e.length;f++){for(var[t,r,i]=e[f],s=!0,u=0;u<t.length;u++)(!1&i||o>=i)&&Object.keys(a.O).every((e=>a.O[e](t[u])))?t.splice(u--,1):(s=!1,i<o&&(o=i));if(s){e.splice(f--,1);var l=r();void 0!==l&&(n=l)}}return n}i=i||0;for(var f=e.length;f>0&&e[f-1][2]>i;f--)e[f]=e[f-1];e[f]=[t,r,i]},a.d=(e,n)=>{for(var t in n)a.o(n,t)&&!a.o(e,t)&&Object.defineProperty(e,t,{enumerable:!0,get:n[t]})},a.f={},a.e=e=>Promise.all(Object.keys(a.f).reduce(((n,t)=>(a.f[t](e,n),n)),[])),a.u=e=>e+".js",a.g=function(){if("object"==typeof globalThis)return globalThis;try{return this||new Function("return this")()}catch(e){if("object"==typeof window)return window}}(),a.o=(e,n)=>Object.prototype.hasOwnProperty.call(e,n),(()=>{var e;a.g.importScripts&&(e=a.g.location+"");var n=a.g.document;if(!e&&n&&(n.currentScript&&(e=n.currentScript.src),!e)){var t=n.getElementsByTagName("script");if(t.length)for(var r=t.length-1;r>-1&&(!e||!/^http(s?):/.test(e));)e=t[r--].src}if(!e)throw new Error("Automatic publicPath is not supported in this browser");e=e.replace(/#.*$/,"").replace(/\?.*$/,"").replace(/\/[^\/]+$/,"/"),a.p=e})(),(()=>{a.b=self.location+"";var e={317:1};a.f.i=(n,t)=>{e[n]||importScripts(a.p+a.u(n))};var n=self.webpackChunkbio=self.webpackChunkbio||[],t=n.push.bind(n);n.push=n=>{var[r,i,o]=n;for(var s in i)a.o(i,s)&&(a.m[s]=i[s]);for(o&&o(a);r.length;)e[r.pop()]=1;t(n)}})(),n=a.x,a.x=()=>a.e(731).then(n);var i=a.x();bio=i})();
-//# sourceMappingURL=317.js.map