# US 2016/0253453 A1 · Parameterizing Cell-to-Cell Regulatory Heterogeneities via Stochastic Transcriptional Profiles

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

- Number: US 2016/0253453 A1
- Document ID: US20160253453A1
- Status: Published application
- Kind: A1
- Application number: 15/000,448
- Filed: 2016-01-19
- Published: 2016-09-01
- Earliest claimed: 2015-01-16
- Assignees: University of Virginia Patent Foundation
- Applicants: University of Virginia Patent Foundation
- Inventors: Kevin JANES; Sameer BAJIKAR; Fabian THEIS; Christiane FUCHS
- Web page: https://patentgrep.com/patent/US20160253453A1

## CPC

- G16B 25/10: Gene or protein expression profiling; Expression-ratio estimation or normalisation
- 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 25/00: ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY

## Abstract

Regulated changes in gene expression underlie many biological processes, but globally profiling cell-to-cell variations in transcriptional regulation is problematic when measuring single cells. Transcriptome-wide identification of regulatory heterogeneities can be robustly achieved by randomly collecting small numbers of cells followed by statistical analysis. However, this stochastic-profiling approach blurs out the expression states of the individual cells in each pooled sample. Various aspects of the disclosure show that the underlying distribution of single-cell regulatory states can be deconvolved from stochastic-profiling data through maximum-likelihood inference. Guided by the mechanisms of transcriptional regulation, the disclosure provides mixture models for cell-to-cell regulatory heterogeneity which result in likelihood functions to infer model parameters. Inferences that validate both computationally and experimentally different mixture models, which include regulatory states for multicellular function occupied by as few as one in 40 cells of the population, are also encompassed. When the disclosed method extends to programs of heterogeneously coexpressed transcripts, the population-level inferences are much more accurate with pooled samples than with one-cell samples when the extent of sampling was limited. The disclosed deconvolution method provides a means to quantify the heterogeneous regulation of molecular states efficiently and gain a deeper understanding of the heterogeneous execution of cell decisions.

## Claims

### Claim 1

A method of inferring single-cell gene expression characteristics, comprising:
a) providing a distribution of expression data of one or more individual genes in more than one sample, wherein each sample comprises more than one cell;
b) providing a model of heterogeneous gene regulation, wherein the model comprises a mixture of two regulatory states; and
c) applying a maximum-likelihood inference to infer the single-cell gene expression characteristics for an individual gene of a).

### Claim 2 (depends on 1)

The method of claim 1, wherein the two regulatory states of an individual gene are characterized as a basal state and an induced regulatory state,
and wherein the basal state and the induced regulatory state can each be approximated by a lognormal distribution,
or wherein the basal state can be approximated by an exponential distribution and the induced regulatory state can be approximated by a lognormal distribution.

### Claim 3 (depends on 2)

The method of claim 2, wherein the basal state and the induced regulatory state can each be approximated by a lognormal distribution,
and wherein the basal state and the induced regulatory state lognormal distributions can be characterized by the following parameters:
μ2 is the log-mean expression of the basal regulatory state;
μ1 is the log-mean expression of the induced regulatory state;
σ is the common log-standard deviation of the basal state and the induced regulatory state;
and F is the probability that a single cell belongs is in the induced regulatory state.

### Claim 4 (depends on 2)

The method of claim 2, wherein the basal state can be approximated by an exponential distribution and the induced regulatory state can be approximated by a lognormal distribution, and wherein the basal state and the induced regulatory state exponential and lognormal distributions can be characterized by the following parameters:
λ is the rate parameter for the basal regulatory state;
μ is the log-mean expression of the induced regulatory state;
σ is the common log-standard deviation of the basal state and the induced regulatory state;
and F is the probability that a single cell belongs is in the induced regulatory state.

### Claim 5 (depends on 3)

The method of claim 3, wherein the maximum-likelihood inference is determined as follows:
fmixture(g)=F·f1(g)+(1−F)·f2(g),
where (g) is a transcript expressed at a low level in the basal regulatory state and at a higher level in the induced regulatory state,
and where f1(g) and f2(g) are defined as:
f
v
(
g
)
,
LN
-
LN

(
x
)
=
1
2


π

σ


x
·
exp

{
-
[
log

(
x
)
-
μ
v
(
g
)
]
2
2


σ
2
}
for


x
>
0


and


v
∈
{
1
,
2
}
,
where the ith random sample of transcript g, Yi(g), is the sum of n independent single-cell expression measurements:
Y
i
(
g
)
=
∑
j
=
1
n

X
ij
(
g
)
,
where Xij(g) is the expression of transcript g in the jth cell of the ith random sample, where the random sample Yi(g) describing the n-cell mixture is defined as:
f
n
L


N
-
L


N

(
y

F
,
μ
1
(
g
)
,
μ
2
(
g
)
,
σ
)
=
∑
j
=
0
n

(
n
j
)

F
j

(
1
-
F
)
n
-
j

f
(
j
,
n
-
j
)
(
g
)
,
L


N
-
L


N

(
y
)
,

where


(
n
j
)


F
j

(
1
-
F
)
n
-
j
is the binomial selection of cells from the basal or induced regulatory states with probabilities F and 1−F, respectively,
where f(j, n−j)(g) is the density of a sum Z1+ . . . +Zn of independent random .variables representing the n-cell draw from the following mixture model:
Z
c
LN
-
LN
=
{
LN

(
μ
1
(
g
)
,
σ
2
)
if


1
≤
c
≤
j
LN

(
μ
2
(
g
)
,
σ
2
)
if


j
<
c
≤
n
and where the probability density function of the sum of lognormally distributed random variables is approximated and applied to the mixture model.

### Claim 6 (depends on 4)

The method of claim 4, wherein the maximum-likelihood inference is determined as follows:
fmixture(g)=F·f1(g)+(1−F)·f2(g),
where (g) is a transcript expressed at a low level in the basal regulatory state and at a higher level in the induced regulatory state,
and where f1(g) and f2(g) are defined as:
f
2
(
g
)
,
EXP
-
LN

(
x
)
=
λ
(
g
)
·
exp

(
-
λ
(
g
)
·
x
)
for


x
≥
0
f
1
(
g
)
,
EXP
-
LN

(
x
)
=
1
2

πσ


x
·
exp

{
-
[
log

(
x
)
-
μ
(
g
)
]
2
2

σ
2
}
for


x
>
0
,
where the ith random sample of transcript g, Yi(g), is the sum of n independent single-cell expression measurements:
Y
i
(
g
)
=
∑
j
=
1
n


X
ij
(
g
)
,
where Xij(g) is the expression of transcript g in the jth cell of the ith random sample, where the random sample Yi(g) describing the n-cell mixture is defined as:
f
n
EXP
-
LN

(
y

F
,
μ
(
g
)
,
λ
(
g
)
,
σ
)
=
∑
j
=
0
n


(
n
j
)

F
j

(
1
-
F
)
n
-
j

f
(
j
,
n
-
j
)
(
g
)
,
EXP
-
LN

(
y
)
,


where


(
n
j
)

F
j

(
1
-
F
)
n
-
j
is the binomial selection of cells from the basal or induced regulatory states with probabilities F and 1−F, respectively,
where f(j, n−j)(g) is the density of a sum Z1+ . . . +Zn of independent random variables representing the n-cell draw from the following mixture model:
Z
c
EXP
-
LN
=
{
LN

(
μ
(
g
)
,
σ
2
)
if


1
≤
c
≤
j
EXP

(
λ
(
g
)
)
if


j
<
c
≤
n
,
where the probability density function of the mixture model is the convolution of a lognormal and an Erlang density.

### Claim 7 (depends on 1)

The method of claim 1, wherein the number of cells is 10.

### Claim 8 (depends on 1)

The method of claim 1, wherein the number of samples is at least 50.

### Claim 9

A method of inferring expression characteristics of a gene cluster, comprising:
a) providing a distribution of expression data of a gene cluster comprising more than one individual gene in more than one sample, wherein each sample comprises more than one cell;
b) providing a model of heterogeneous gene regulation, wherein the model comprises a mixture of two regulatory states; and
c) applying a maximum-likelihood inference to infer the expression characteristics of the gene cluster.

### Claim 10 (depends on 9)

The method of claim 9, wherein the two regulatory states of the individual gene within the gene cluster are characterized as a basal state and an induced regulatory state,
and wherein the basal state and the induced regulatory state can each be approximated by a lognormal distribution,
or wherein the basal state can be approximated by an exponential distribution and the induced regulatory state can be approximated by a lognormal distribution.

### Claim 11 (depends on 10)

The method of claim 10, wherein the basal state and the induced regulatory state can each be approximated by a lognormal distribution,
wherein individual genes are characterized by the following parameters;
μ2 is the log-mean expression of an individual gene in the basal regulatory state, and
μ2 is the log-mean expression of an individual gene in the induced regulatory state;
and wherein the basal state and the induced regulatory state lognormal distributions can be characterized by the following parameters:
σ is the common log-standard deviation of the basal state and the induced regulatory state, and
F is the probability that a single cell belongs is in the induced regulatory state for all genes within the gene cluster,
wherein the individual genes of the gene cluster share a common σ and F.

### Claim 12 (depends on 10)

The method of claim 10, wherein the basal state can be approximated by an exponential distribution and the induced regulatory state can be approximated by a lognormal distribution, and wherein the basal state and the induced regulatory state exponential and lognormal distributions can be characterized by the following parameters:
λ is the rate parameter of an individual gene in the basal regulatory state;
μ is the log-mean expression of an individual gene in the induced regulatory state;
σ is the common log-standard deviation of the basal state and the induced regulatory state;
and F is the probability that a single cell belongs is in the induced regulatory state.

### Claim 13 (depends on 11)

The method of claim 11, wherein the maximum-likelihood inference is determined as follows:
fmixture(g)=F·f1(g)+(1−F)·f2(g),
where (g) is a transcript expressed at a low level in the basal regulatory state and at a higher level in the induced regulatory state,
and where f1(g) and f2(g) are defined as:
f
v
(
g
)
,
LN
-
LN

(
x
)
=
1
2

π

σ


x
·
exp

{
-
[
log

(
x
)
-
μ
v
(
g
)
]
2
2

σ
2
}

for



x
>
0


and


v
∈
{
1
,
2
}
,
where the ith random sample of transcript g, Yi(g), is the sum of n independent single-cell expression measurements:
Y
i
(
g
)
=
∑
j
=
1
n


X
ij
(
g
)
,
where Xij(g) is the expression of transcript g in the jth cell of the ith random sample,
where the random sample Yi(g) describing the n-cell mixture has the pdf:
f
n
LN
-
LN

(
y

F
,
μ
1
(
g
)
,
μ
2
(
g
)
,
σ
)
=
∑
j
=
0
n


(
n
j
)

F
j

(
1
-
F
)
n
-
j

f
(
j
,
n
-
j
)
(
g
)
,
LN
-
LN

(
y
)
,


where


(
n
j
)

F
j

(
1
-
F
)
n
-
j
is the binomial selection of cells from the basal or induced regulatory states with probabilities F and 1−F, respectively,
where f(j, n−j)(g) is the density of a sum Z1+ . . . +Zn of independent random variables representing the n-cell draw from the following mixture model:
Z
c
LN
-
LN
=
{
LN

(
μ
1
(
g
)
,
σ
2
)
if


1
≤
c
≤
j
LN

(
μ
2
(
g
)
,
σ
2
)
if


j
<
c
≤
n
,
where the probability density function of the sum of lognormally distributed random variables is approximated and applied to the mixture model,
and where the log-likelihood function for the model parameters given k random n-cell samples is:

LN
-
LN

(
F
,
μ
1
_
,
μ
2
_
,
σ
)
=
∑
g
=
1
m


∑
i
=
1
k


log

[
f
n
LN
-
LN

(
y
i
(
g
)

F
,
μ
i
(
g
)
,
μ
2
(
g
)
,
σ
)
]
,
where the gene cluster is m transcripts,
where μ1 and μ2, are vectors containing the transcript-specific log-means for the two regulatory states: μ1=(μ1(l), . . . , μ1(m)) and μ2=(μ2(l), . . . , μ2(m)),
and where the log-likelihood function assumes that the expression levels of each transcript are independent as defined by the specific mixture model and F.

### Claim 14 (depends on 13)

The method of claim 13, wherein the gene cluster is broken down into four-gene subgroups to infer log-means for each gene in the subgroup and local estimates of F and σ.

### Claim 15 (depends on 12)

The method of claim 12, wherein the maximum-likelihood inference is determined as follows:
the single-cell expression for transcript g follows the probability density function:
fmixture(g)=F·f1(g)+(1−F)·f2(g),
where (g) is a transcript expressed at a low level in the basal regulatory state and at a higher level in the induced regulatory state,
and where f1(g) and f2(g) are defined as:
f
2
(
g
)
,
EXP
-
LN

(
x
)
=
λ
(
g
)
·
exp

(
-
λ
(
g
)
·
x
)
for


x
≥
0
f
1
(
g
)
,
EXP
-
LN

(
x
)
=
1
2

πσ


x
·
exp

{
-
[
log

(
x
)
-
μ
(
g
)
]
2
2

σ
2
}
for


x
>
0
,
where the ith random sample of transcript g, Yi(g), is the sum of n independent single-cell expression measurements:
Y
i
(
g
)
=
∑
j
=
1
n


X
ij
(
g
)
,
where Xij(g) is the expression of transcript g in the jth cell of the ith random sample,
where the random sample Yi(g) describing the n-cell mixture has the probability density function:
f
n
EXP
-
LN

(
y

F
,
μ
(
g
)
,
λ
(
g
)
,
σ
)
=
∑
j
=
0
n


(
n
j
)

F
j

(
1
-
F
)
n
-
j

f
(
j
,
n
-
j
)
(
g
)
,
EXP
-
LN

(
y
)
,


where


(
n
j
)

F
j

(
1
-
F
)
n
-
j
is the binomial selection of cells from the basal or induced regulatory states with probabilities F and 1−F, respectively,
where f(j, n−j)(g) is the density of a sum Z1+ . . . +Zn of independent random variables representing the n-cell draw from the following mixture model.
Z
c
EXP
-
LN
=
{
LN

(
μ
(
g
)
,
σ
2
)
if


1
≤
c
≤
j
LN

(
λ
(
g
)
)
if


j
<
c
≤
n
,
where the probability density function of the mixture model is the convolution of a lognormal and an Erlang density,
and where the log-likelihood function for the model parameters given k random n-cell samples is:

EXP
-
LN

(
F
,
μ
_
,
λ
_
,
σ
)
=
∑
g
=
1
m


∑
i
=
1
k


log

[
f
n
EXP
-
LN

(
y
i
(
g
)

F
,
μ
(
g
)
,
λ
(
g
)
,
σ
)
]
,
where the gene cluster is m transcripts,
where μ, and λ are vectors containing the transcript-specific log-means for the two regulatory states: μ=(μ(l), . . . , μ(m)), and λ=(λ(l), . . . λ(m)),
and where the log-likelihood function assumes that the expression levels of each transcript are independent as defined by the specific mixture model and F.

### Claim 16 (depends on 15)

The method of claim 15, wherein the gene cluster is broken down into four-gene subgroups to infer log-means for each gene in the subgroup and local estimates of F, σ, and λ.

### Claim 17 (depends on 9)

The method of claim 9, wherein the number of cells is 10.

### Claim 18 (depends on 9)

The method of claim 9, wherein the number of samples is less than 50.

### Claim 19 (depends on 5)

The method of claim 5, wherein log-standard deviations of the basal state and the induced regulatory state are individually characterized as σ2 and σ1.

### Claim 20 (depends on 13)

The method of claim 13, wherein log-standard deviations of the basal state and the induced regulatory state are individually characterized as σ2 and σ1.
