Namespace: machineLearning

machineLearning

Source:

Methods

(static) doubleClustering(matrix, rowNames, colNames, metricopt) → {Object}

Performs double clustering on a matrix.
Parameters:
Name Type Attributes Default Description
matrix Array.<Array.<number>> The matrix to be clustered.
rowNames Array.<string> The names of the rows in the matrix.
colNames Array.<string> The names of the columns in the matrix.
metric function <optional>
 euclideanDistance The distance metric to use for clustering.
Source:
Returns:
An object with three properties: matrix, rowNames, and colNames. matrix is the clustered matrix. rowNames and colNames are the names of the rows and columns in the clustered matrix, respectively.
Type
Object

(static) exports.kFoldCV(Xs, Ys, kopt, modelTypeopt) → {Object}

Performs k-fold stratified cross-validation for multivariate linear or MLP regression models.
Parameters:
Name Type Attributes Default Description
Xs Array Array of input data.
Ys Array Array of output data.
k number <optional>
 10 Number of folds for cross-validation.
modelType string <optional>
 "MLR" Regression model type ("MLR" for multivariate linear regression or "MLP" for multilayer perceptron).
Source:
Throws:
- If an unknown model type is provided.
Type
Error
Returns:
- Object containing an array of trained regression models, an array of mean squared errors for each fold, and the average mean squared error across all folds.
Type
Object

(static) exports.preprocessData(mutationalData, exposureData, dataSource) → {Object}

Preprocesses mutational and exposure data for a given data source. Currently only supports data from MSigDB portal.
Parameters:
Name Type Description
mutationalData Array Array of mutational data.
exposureData Array Array of exposure data.
dataSource string Data source identifier.
Source:
Throws:
- If an unknown data source is provided.
Type
Error
Returns:
- Object containing input (Xs) and output (Ys) arrays for regression.
Type
Object

(async, static) fitMutationalSpectraToSignatures(mutationalSignatures, mutationalSpectra, optionsopt) → {Object}

Fits mutational spectra to known mutational signatures using non-negative least squares (NNLS). This function calculates the exposure of mutational signatures for each sample by fitting the observed mutational spectra to the reference mutational signatures. It then filters out signatures whose contribution is below a fraction of the total exposure, by setting their exposures to zero. You can choose to return exposures as absolute values (raw counts from NNLS) or relative values (sum to 1 for each sample). The filtering threshold is a fraction between 0 and 1.
Parameters:
Name Type Attributes Description
mutationalSignatures Object Reference mutational signatures. Each key is a signature name, and each value is an object of mutation types (e.g., {"C>A": weight, "C>G": weight}).
mutationalSpectra Object Mutational spectra for each sample. Each key is a sample ID, and each value is an object of mutation types with their counts (e.g., {"C>A": count, "C>G": count}).
options Object <optional>
 Configuration options for filtering and output.
Properties
Name Type Attributes Default Description
exposureThreshold number <optional>
 0 Exclude signatures below this fraction of the total, by setting their exposures to zero. Must be between 0 and 1.
exposureType "absolute" | "relative" <optional>
 "relative" Return exposures as absolute or relative.
renormalize boolean <optional>
 true Whether to normalize exposures so that they sum to 1 after filtering.
Source:
Returns:
- An object with sample IDs as keys. Each value is an object of signature exposures.
Type
Object
Example
// Example usage:
// 1. Get mutational signatures (e.g., from the mSigPortal API)
const mutationalSignatures = await mSigPortal.mSigPortalData.getMutationalSignaturesData(
  "WGS", "COSMIC_v3_Signatures_GRCh37_SBS96", "SBS", 96, 1000
);

// 2. Extract mutational spectra for each sample
const extractedSpectra = await mSigPortal.mSigPortalData.extractMutationalSpectra(
  mutationalSignatures, "signatureName"
);

// 3. Fit spectra to signatures with post-fit filtering
const nnlsExposures = await mSigPortal.signatureFitting.fitMutationalSpectraToSignatures(
  mutationalSignatures,
  extractedSpectra,
  {
    exposureThreshold: 0.1,
    exposureType: "relative",
    renormalize: true
  }
);

console.log(nnlsExposures);
// {
//   Sample1: { SBS1: 0.75, SBS2: 0.25, SBS3: 0 },
//   Sample2: { SBS1: 0.9, SBS2: 0.1, SBS3: 0 },
//   ...
// }

(static) hierarchicalClustering(distanceMatrix, sampleNames) → {Object}

Performs hierarchical clustering on a distance matrix.
Parameters:
Name Type Description
distanceMatrix Array.<Array.<number>> The distance matrix to be clustered.
sampleNames Array.<string> The names of the samples in the distance matrix.
Source:
Returns:
The final clustering result as a tree.
Type
Object

(async, static) nnls(A, b, maxiteropt) → {Object}

Solves argmin_x || Ax - b ||_2 for x>=0. A is a matrix, b is a vector.
Parameters:
Name Type Attributes Default Description
A Array.<Array.<number>> The matrix A.
b Array.<number> The vector b.
maxiter number <optional>
 3*A[0].length Maximum number of iterations.
Source:
Returns:
An object with two properties: x and rnorm. x is a vector with the same length as b. rnorm is the residual || Ax - b ||^2.
Type
Object

(static) upgma(distanceMatrix) → {Array}

Performs UPGMA clustering on a distance matrix.
Parameters:
Name Type Description
distanceMatrix Array.<Array.<number>> The distance matrix to be clustered.
Source:
Returns:
An array of arrays representing the clustering result. Each inner array contains three elements: two clusters being merged and their average distance.
Type
Array