Spanning-tree progression analysis of density-normalized events (SPADE)

• US 10,289,802 B2
• Filed: 03/02/2012
• Issued: 05/14/2019
• Est. Priority Date: 12/27/2010
• Status: Active Grant

• Current Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford Management Co.)
• Original Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford Management Co.)
• Inventors: Qiu, Peng, Gentles, Andrew J., Plevritis, Sylvia K., Nolan, Garry, Sachs, Karen
• Primary Examiner(s): Clow, Lori A.
• Application Number:
  US13/411,492
• Publication Number:
  US 20130060775A1
• Time in Patent Office: 2,629 Days
• Field of Search: None
• CPC Class Codes: G16B 25/00 ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
• CPC Class Codes: G16B 40/00 ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• CPC Class Codes: G16B 40/30 Unsupervised data analysis
• CPC Class Codes: G16B 45/00 ICT specially adapted for bioinformatics-related data visualisation, e.g. displaying of maps or networks

Abstract

Methods and systems for determining progression and other characteristics of microarray expression levels and similar information, alternatively using a network or communications medium or tangible storage medium or logic processor.

Claims

1. A computer implemented method for clustering and visualization of multicolor flow cytometry data, the method comprising:

receiving cell samples from at least one subject;

analyzing the cell samples using a flow cytometer, thereby yielding a multi-dimensional data set comprising cell sample points, wherein each cell sample point provides information about a measurement of a feature of a single cell sample;

estimating a density function for the cell sample points in the multi-dimensional data set using a computing device;

creating a down-sampled data set by removing a portion of the cell sample points in dense regions of the multi-dimensional data set using the computing device, wherein the dense regions are determined by the density function, wherein said creating comprises;

determining a local density for each cell sample in said multi-dimensional data set using said computing device;

downsampling cell sample points with local densities above a target density, so that their local densities reduce to approximately the target density after downsampling; and

preventing downsampling of cell sample points with local densities less than said target density;

clustering the cell sample points in the down-sampled data set into a plurality of sample clusters using the computing device;

creating at least one progression tree in a Euclidean space using the computing device, wherein the at least one progression tree represents a progression among the plurality of sample clusters and reveals features associated with the progression;

rendering an image using the created at least one progression tree using the computing device, wherein the image graphically shows a progressive relationship underlying the plurality of sample clusters using the computing device;

receiving user inputs for the rendered image to interactively select a set of progressively related portions of the plurality of sample clusters;

gating the multi-dimensional data set based on the selected sets of portions; and

interactively rendering and displaying the gated multi-dimensional data set in response to received user inputs using the computing device.

2. The method of claim 1 further wherein:

the multi-dimensional data set comprises an array of M samples by N features;

the down-sampled data set comprises a down-sampled array of M* samples by N* features;

the plurality of sample clusters comprise C clusters.

3. The method of claim 1, wherein rendering the image comprises:

creating a visual representation of the at least one progression tree, wherein the visual representation shows data indicating intensity of at least one of the features.

4. The method of claim 1 further comprising:

storing additional numerical feature values in the multi-dimensional data set, the additional numerical feature values not used in creating the down-sampled data set or clustering the cell sample points;

wherein rendering the image comprises creating a representation of the at least one progression tree, the representation comprising a plurality of nodes and a plurality of edges connecting the plurality of nodes, wherein each of the plurality of edge includes a distance values, wherein a portion of the representation is in a color that is different from the remaining portion of the representation, the distance values and connection of nodes indicating a progression of the cell samples points using said features, and color or other indicators of said nodes or said tree indicating at least one additional numerical feature value.

5. The method of claim 1 further comprising:

selecting a first subset of said features to use for said creating the down-sampled data set;

selecting a second subset of said features to use for said clustering; and

selecting a third subset of said features to use for said creating the at least one progression tree.

6. The method of claim 1 further wherein said cell samples comprise a plurality of cells and said features comprise a number of detectable cellular features or characteristics, the method further providing:

constructing and outputting progression trees based on a selected subset of features to perform gating.

7. The method of claim 1 further wherein:

abundant and rare cell sample types are relatively equally represented, and rare sample types are more likely to form their own clusters; and

the number of cell sample points in said multi-dimensional data set is significantly reduced while most cell sample points of the rare sample type remain after downsampling and an overall distribution or shape of cell samples in the array of the original dataset is preserved.

8. The method of claim 1 further comprising:

after said clustering is complete, said computing device upsamples the cluster data by assigning each cell sample in the original dataset to a cluster and said upsampling comprises;

calculating median intensity and other statistics of each cluster with high accuracy;

assigning each cell sample in the original dataset to one cluster by determining its nearest neighbor in the down-sampled data set and assigning the sample to the cluster that the nearest neighbor in the downsampled data belongs to.

9. The method of claim 1 further wherein creating the progression tree comprises:

defining a fully connected undirected weighted graph comprising a plurality of nodes and a plurality of edges connecting the nodes, wherein each node represents one sample cluster;

determining a weight on the edge that connects nodes i and j is defined as the Euclidean distance between the feature expressions of sample clusters i and j;

applying Boruvka'"'"'s algorithm to derive the MST from the fully connected graph, such that starting from one cluster and moving along the edges of the MST, a gradual change of feature expression levels is observed.

10. The method of claim 1 further comprising:

analyzing dynamics of features under different perturbations;

such that for a feature and a perturbation, determining for each cluster a ratio between the median intensities of features in an unstimulated condition and the median intensities of features in a stimulated condition;

indicating dynamics of features under perturbations in said outputting of cluster representations in said trees using said ratio.

11. The method of claim 1 further comprising:

determining from said progression tree, one or more progression branches;

comparing said progression branches across multiple features; and

associating said branches with one or more progression states of said samples.

12. The method of claim 1 further wherein said samples comprise a plurality of cells and said features comprise a number of detectable cellular features or characteristics, the method further providing identifying uncharacterized features by correlating their progression behavior to well-characterized features.

13. The method of claim 1 further comprising:

determining a progression similarity matrix between two or more features indicating a number of progressions supported in common by the feature;

wherein the progression similarity matrix is an integer count of progression trees concordant with the feature according to a selected threshold.

14. The method of claim 13 further wherein the progression similarity matrix quantifies the progression similarity between pairs of features, wherein the u;

v element of the progression similarity matrix is a number of progression trees that are concordant with both features u and v.

15. The method of claim 1 further comprising:

receiving into a provided memory readable and writable by a provide CPU flow cytometry data, said flow cytometry data containing data values indicative of plurality of features measured for a larger plurality of samples;

clustering samples into a smaller number of sample clusters, where clusters are determined by comparing features across multiple samples from the received flow cytometry data;

determining per-feature progressions for one or more selected features from the flow cytometry data;

identifying progression-similar features by identifying which features have high progression similarity to multiple per-feature progressions;

using the progression-concordant features to determine a most likely overall progression of samples from flow cytometry data; and

outputting said most likely overall progression of flow cytometry samples to a user using a provided computer output device.