# US 2010/0144554 A1 · Methods for assembling panels of cancer cell lines for use in testing the efficacy of one or more pharmaceutical compositions

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2010/0144554 A1
- Document ID: US20100144554A1
- Status: Published application
- Kind: A1
- Application number: 12/607,077
- Filed: 2009-10-28
- Published: 2010-06-10
- Earliest claimed: 2008-10-31
- Assignees: Abbott Laboratories
- Inventors: Dimitri Semizarov; Xin Lu; Ke Zhang; Rick R. Lesniewski
- Web page: https://patentgrep.com/patent/US20100144554A1

## CPC

- G16B 40/30: Unsupervised data analysis
- G16B 5/00: ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
- G16B 5/20: Probabilistic models
- G16B 20/10: Ploidy or copy number detection
- G16B 20/20: Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
- 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
- G16H 50/50: for simulation or modelling of medical disorders
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G06F 17/10: Complex mathematical operations {(function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544)}
- G06F 17/11: for solving equations {, e.g. nonlinear equations, general mathematical optimization problems (optimization specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06F 17/15: Correlation function computation {including computation of convolution operations (arithmetic circuits for sum of products per se, e.g. multiply-accumulators G06F7/5443; digital filters, e.g. FIR, IIR, adaptive filters H03H17/00)}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F 17/17: Function evaluation by approximation methods, e.g. inter- or extrapolation, smoothing, least mean square method ({G06F17/18 takes precedence } ; interpolation for numerical control G05B19/18)
- G16B 20/00: ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

The present invention relates to algorithms for use in defining genomic subgroups of tumors and cancer cell lines. The present invention also relates to methods for assembling panels of tumors and cancer cell lines according to genomic subgroups for use in testing the efficacy of one or more pharmaceutical compounds in the treatment of subjects suffering from at least one cancer.

## Claims

### Claim 1

An algorithm for use in clustering tumors and cell lines to define genomic subgroups, the method comprising the steps of:
(a) obtaining a plurality of m samples comprising at least one tumor or cancer cell line;
(b) acquiring a data set comprising copy number alteration information from at least one locus from each chromosome from each sample obtained in step (a);
(c) identifying in the data set, copy number alteration information obtained from samples contaminated by normal cells and eliminating the contaminated samples from the data set, wherein the identifying and eliminating comprises:
(1) applying a machine learning algorithm tuned to parameters that represent the differences between tumor and normal samples to the data;
(2) assigning a probability score for normal cell contamination to each sample as determined by the machine learning algorithm;
(3) eliminating data from the data set for each sample scoring 50% or greater probability of containing normal cells;
(d) estimating a number of subgroups, r, in the data set by applying an unsupervised clustering algorithm using Pearson linear dissimilarity algorithm to the data set;
(e) assigning each sample in the data set to at least one cluster using a modified genomic non-negative matrix factorization (gNMF) algorithm, wherein the modified gNMF algorithm comprises:
(1) calculating divergence of the algorithm after every 100 steps of multiplicative updating using the formula (1):
D
(
V


WH
)
=
∑
i
=
1
n

∑
j
=
1
m

(
V
ij

log


V
ij
(
WH
)
ij
-
V
ij
+
(
WH
)
ij
)
(
1
)
wherein the Vij is the ith row and jth column of matrix V, (WH)ij is the ith row and jth column of matrix (W*H), i runs from 1 to n and n is the number of segments in the data set, and j runs from 1 to m and m is the number of samples in the data set.
(2) stopping the algorithm if the divergence calculated in step (e) (1) does not decrease by more than about 0.001% when compared to the divergence calculated for the previous 100 steps of multiplicative updating of the algorithm;
(3) randomly repeating the algorithm for a selected number of runs and calculating a Pearson correlation coefficient matrix of H for the each of run the algorithm using the formula (2):
C
i
,
j
=
ρ

(
H
,
i
,
H
,
j
)
=
1
r
-
1

∑
k

(
H
k
,
i
-
H
,
j
_
)

(
H
k
,
j
-
H
,
j
_
)
s
H
,
i

s
H
,
j
(
2
)
herein C is the correlation matrix, Ci,j is the ith row and jth column in the matrix C, H,i and H,j are the ith and jth column vector in matrix H, ρ(H,i, H,j) is the Pearson correlation coefficient between H,i and H,j, i and j run from 1 to m and m is the number of samples in the data set, k runs from 1 to r and r is the number of subgroups from step (d);
(4) averaging the Pearson correlation coefficient matrices for each run of the algorithm obtained from step (e)(3) to arrive at an average correlation matrix;
(5) assigning tumors and cancer cell lines in the data set into r subgroups by applying a unsupervised clustering algorithm using 1 minus the average correlation matrix determined in step (e)(4) and cutting a dendrogram into r clusters;
(f) applying a Cophenetic correlation, Bayesian Information Criterion or a combination thereof to provide a final number of clusters from the data set, wherein each final cluster defines a genomic subgroup for each tumor or cancer cell line sample; and
(g) evaluating the stability of the final number of clusters selected in step (f) using a ten-fold stability test.

### Claim 2 (depends on 1)

The algorithm of claim 1, wherein the unsupervised clustering algorithm is a hierarchical clustering.

### Claim 3 (depends on 1)

The algorithm of claim 1, wherein Cophenetic correlation is used to provide a final number of clusters from the data set.

### Claim 4 (depends on 1)

The algorithm of claim 1, wherein Bayesian Information Criterion is used to provide a final number of clusters from the data set.

### Claim 5 (depends on 1)

The algorithm of claim 1, wherein Cophenetic correlation and Bayesian Information Criterion are used to provide a final number of clusters from the data set.

### Claim 6

A method for assembling panels of tumor and cancer cell lines according to genomic subgroups, the method comprising the steps of:
(a) obtaining a plurality of m samples comprising at least one tumor or cancer cell line;
(b) acquiring a data set comprising copy number alteration information from at least one locus from each chromosome from each sample obtained in step (a);
(c) identifying in the data set, copy number alteration information obtained from samples contaminated by normal cells and eliminating the contaminated samples from the data set, wherein the identifying and eliminating comprises:
(1) applying a machine learning algorithm tuned to parameters that represent the differences between tumor and normal samples to the data;
(2) assigning a probability score for normal cell contamination to each sample, as determined by the machine learning algorithm;
(3) eliminating data from the data set for each sample scoring 50% or greater probability of containing normal cells;
(d) estimating a number of subgroups, r, in the data set by applying unsupervised clustering using Pearson linear dissimilarity algorithm to the data set;
(e) assigning each sample in the data set to at least one cluster using a modified genomic non-negative matrix factorization (gNMF) algorithm, wherein the modified gNMF algorithm comprises:
(1) calculating divergence of the algorithm after every 100 steps of multiplicative updating using the formula (1):
D
(
V


WH
)
=
∑
i
=
1
n

∑
j
=
1
m

(
V
ij

log


V
ij
(
WH
)
ij
-
V
ij
+
(
WH
)
ij
)
(
1
)
wherein the Vij is the ith row and jth column of matrix V, (WH)ij is the ith row and jth column of matrix (W*H), i runs from 1 to n and n is the number of segments in the data set, and j runs from 1 to m and m is the number of samples in the data set.
(2) stopping the algorithm if the divergence calculated in step (e) (1) does not decrease more than about 0.001% when compared to the divergence calculated for the previous 100 steps of multiplicative updating of the algorithm;
(3) randomly repeating the algorithm for a selected number of runs and calculating a Pearson correlation coefficient matrix of H for the each of run the algorithm using the formula (2):
C
i
,
j
=
ρ

(
H
,
i
,
H
,
j
)
=
1
r
-
1

∑
k

(
H
k
,
i
-
H
,
j
_
)

(
H
k
,
j
-
H
,
j
_
)
s
H
,
i

s
H
,
j
(
2
)
herein C is the correlation matrix, Ci,j is the ith row and jth column in the matrix C, H,i and H,j are the ith and ith column vector in matrix H, ρ(H,i, H,j) is the Pearson correlation coefficient between H,i and H,j, i and j run from 1 to m and m is the number of samples in the data set, k runs from 1 to r and r is the number of subgroups from step (d);
(4) averaging the Pearson correlation coefficient matrices for each run of the algorithm obtained from step (e)(3) to arrive at an average correlation matrix;
(5) assigning tumors and cancer cell lines in the data set into r subgroups by applying a unsupervised clustering algorithm using 1 minus the average correlation matrix determined in step (e)(4) and cutting a dendrogram into r clusters;
(f) applying a Cophenetic correlation, Bayesian Information Criterion or a combination thereof to select a final number of clusters from the data set;
(g) evaluating the stability of the final number of clusters selected in step (f) using a ten-fold stability test; and
(h) selecting at least one tumor or cell line from each cluster selected in step (f) and assembling the selected cell lines into panels defined according to genomic subgroups.

### Claim 7 (depends on 6)

The method of claim 6, wherein the cancer is selected from the group consisting of: small cell lung carcinoma, non-small cell lung carcinoma, colorectal cancer, and melanoma.

### Claim 8 (depends on 6)

The method of claim 6, wherein the copy number alteration is a gain or loss or copy number.
