# Bayesian Functional Data Clustering for Temporal Microarray Data

^{1}Department of Statistics, University of Illinois, Champaign, IL 61820, USA

^{2}Department of Statistics, Harvard University, Cambridge, MA 02138, USA

This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

## Abstract

We propose a Bayesian procedure to cluster temporal gene expression microarray profiles, based on a mixed-effect smoothing-spline model, and design a Gibbs sampler to sample from the desired posterior distribution. Our method can determine the cluster number automatically based on the Bayesian information criterion, and handle missing data easily. When applied to a microarray dataset on the budding yeast, our clustering algorithm provides biologically meaningful gene clusters according to a functional enrichment analysis.

## 1. INTRODUCTION

Microarray technology enables the scientist to measure the mRNA expression levels of thousands of genes simultaneously. For a particular species of interest, one can make microarray measurements under many different conditions and for different types of cells (if it is a multicellular organism). Genes' expression profiles under these conditions often give the scientist some clues on biological roles of these genes. A group of genes with similar profiles are often “coregulated” or participants of the same biological functions.

When a series of microarray experiments are conducted sequentially during a biological process, we call the resulting dataset a “temporal” microarray dataset, which can provide insights on the underlying biology and help decipher the dynamic gene regulatory network. Clustering genes with similar temporal profiles is a crucial first step to reveal potential relationships among the genes.

Conventional clustering methods, such as the K-means and hierarchical clustering, do not take into consideration the correlation in the gene expression levels over time. Although it is possible to use a general multivariate Gaussian model to account for the correlation structure, such a model ignores the time order of the gene expressions. As evidenced in our example, the time factor is important in interpreting the results of gene expression clustering in temporal data. It is also possible to use an autoregression model to describe the gene expression time series, but such a model often requires stationarity, which is unlikely to hold in most temporal microarray data.

Recently, nonparametric analysis of data in the form of curves, that is, functional data, is subject to active research, see [1, 2] for a comprehensive treatment of functional data analysis; and curve-based functional clustering methods have emerged [3–7], but a rigorous assessment of the estimation precision is still lacking.

In this paper, we propose a Bayesian clustering method, which optimally combines the available information and provides a proper uncertainty measure for all estimated quantities. Our method is based on a mixture of mixed-effect smoothing splines models. For each cluster, we model its mean profile as a smoothing spline function and describe its individual gene's variation by a parametric random effect. Based on the theory of reproducing-kernel Hilbert spaces [8], we represent the mean expression curve as a linear combination of certain basis functions, which enables us to derive the full posterior distribution up to a normalizing constant. All the conditional distributions needed by a Gibbs sampler are also easy to compute and to sample from. Our method automatically takes care of the missing data and infers the number of clusters in the data. Using the method, we analyzed a microarray dataset of budding yeast, we found that the majority of the clusters we had obtained are enriched for known and expected biological functions.

Our method is not restricted to temporal microarray data, and can be applied to all curve clustering problems, especially for sparsely and irregularly sampled temporal data.

## 2. MATERIAL AND METHODS

### 2.1. Mixed-effect representation of gene expression profile

Let the expression value of the *i*th gene at time *t* be *y _{it}*. To accommodate missing data that occasionally occurs
in microarray experiment, we denote

**t**

_{i}= (

*t*

_{1},…,

*t*

_{ni}) and

**y**

_{i}= (

*y*

_{i1},…,

*y*)

_{ini}^{T}, where

*n*is the number of measurements of

_{i}*i*th gene. Our mixed-effect smoothing spline model [9] for genes in one cluster is

where * μ*(

**t**

_{i}) = (

*μ*(

*t*

_{1}),…

*μ*(

*t*))

_{ni}^{T}is the cluster's mean profile,

**b**

_{i}∼

*N*(0,

*B*) is the random effect to capture the intragene correlation,

*Z*is the known design matrix for the random effect, and

_{i}*ϵ*∼

_{i}*N*(0,

*σ*

^{2}

*I*) is the random error independent of

**b**and of each other.

By taking different **b** vectors, we can
accommodate different nonrandom effects. For example, when **b**
_{i} = *b _{i}* and

*Z*=

_{i}**1**, the expression profile of the

*i*th gene is parallel to the mean profile

**(Figure 1). If**

*μ***b**

_{i}= (

*b*

_{i1},

*b*

_{i2})

^{T}and

*Z*= (

_{i}**1**,

**t**

_{i}), the difference between the

*i*th gene profile and the mean profile is a linear function in time. More complicated structures such as periodicity can be modeled by letting the

*Z*be basis of a certain functional space.

_{i}By considering *μ* in a
reproducing kernel Hilbert space *ℋ* ⊆ {*μ* : *M*(*μ*) < ∞} in which *M*(*μ*) is a square
seminorm, we can represent *μ* as

where {*s _{j}*} is a set
consisting of all distinct {

*t*},

_{i}*q*is the number of {

*s*}, and

_{j}*R*is the kernel of

_{M}*ℋ*. The choice of

*M*(

*μ*) = ∫

^{a}

_{0}(

*d*

^{2}

*μ*/

*dt*

^{2})

^{2}

*dt*yields the cubic smoothing spline with

where (·)_{+} = max(·, 0) [10].

Writing (2) in a vector-matrix form, we have

where *S _{i}* is

*n*×

_{i}*m*with the (

*i*,

*ν*)th entry

*ϕ*

_{ν}(

*t*) and

_{i}*R*is

*n*×

_{i}*q*with the (

*i*,

*j*)th entry

*R*(

_{M}*t*,

_{i}*s*). Substituting (5) into (1), we have

_{j}
Denoting **y** = (**y**
^{T}
_{1},…, **y**
^{T}
_{n})^{T} and *S*, *R*, *Z*, * ϵ* similarly, we
have the matrix representation

where **b** = (**b**
^{T}
_{1},…, **b**
^{T}
_{n})^{T} ∼ *N*(0, diag(*B*,…, *B*)).

The prior distributions are specified as follows:

where IG and IW are inverse-Gamma and inverse-Wishart distributions, respectively.

These priors lead to the following full conditional posteriors, which are used in our Gibbs sampler:

where *V _{d}* = (

*S*

^{T}*S*/

*σ*

^{2}+ diag(

*δ*

^{−1}

_{1},…,

*δ*

^{−1}

_{m}))

^{−1},

*V*= (

_{b}*Z*

^{T}*Z*/

*σ*

^{2}+ diag(

*B*

^{−1},…,

*B*

^{−1}))

^{−1},

*V*= (

_{c}*R*

^{T}*R*/

*σ*

^{2}+ 1/

*τ*

^{2}

*I*)

^{−1}, and SSR = (

**y**−

*S*

**d**−

*R*

**c**−

*Z*

**b**)

^{T}(

**y**−

*S*

**d**−

*R*

**c**−

*Z*

**b**).

### 2.2. The mixture model with unknown number of components

When more than one cluster is considered, we assume
that the expression of the *i*th gene has a
Gaussian mixture distribution:

where * μ_{k}* and Σ

_{k}=

*ZB*+

_{k}Z^{T}*σ*

^{2}

*I*are the mean and covariance matrix for the

*k*th component, as given by (7);

*p*

_{k}is the fraction of

*k*th component, and

*K*is the number of Gaussian components.

### 2.3. Class labels and cluster numbers

To ease the
computation, we introduce a “latent” membership labeling variable *j _{i}* for the

*i*th gene so that

When the number of Gaussian
components *K* is known, we
can get the joint posterior probability as

where **J** = (*j*
_{1},…, *j*
_{n}), * μ* = (

*μ*

_{1},…,

*μ*

_{K}), Σ = (Σ

_{1},…, Σ

_{K}), and

*π*(

*, Σ) is the joint prior distribution.*

**μ**Since *K* is unknown, we
used the following Bayesian information criterion (BIC):

where *M _{K}* is the current
model with parameters

*, ${\widehat{\mathbf{\theta}}}_{K}$ is the estimate, and*

**θ**_{K}*l*is the total number of parameters in our model. A small BIC score indicates the adequacy of the corresponding model. An alternative to our current approach (i.e., each clustering configuration is equally likely given the number of clusters

_{K}*K*, and

*K*is determined by BIC) is to use a Polya Urn prior (also called the “Chinese restaurant” process), which postulates that when a new member comes in, its a priori probability for joining an existing cluster of size

*m*is (

_{i}*m*+

_{i}*c*)/(

*m*+

*c*), and for forming a new cluster of its own is

*c*/(

*m*+

*c*), where

*m*is the total number of existing members. This prior, however, favors unbalanced cluster configurations (e.g., very large and very small clusters) and may not be appropriate in our applications.

#### 2.3.1. Gibbs Sampling from the Posterior

To complete our Bayesian analysis, we employ the
Dirichlet prior Di (*α*
_{1},…, *α*
_{K}) for (*p*
_{1},…, *p*
_{K}), the cluster proportions. Thus, given the cluster
indicator **J**, the posterior distribution of the *p*'s is again a
Dirichlet distribution.

Given **μ**_{1},…, **μ**_{K}, *B*
_{1},…, *B*
_{K}, *σ*
^{2}, we have the
conditional distribution of *j _{i}*:

With an initial value of **J**, which gives rise to a partition of **y** : (**y**
^{J}
_{1},…, **y**
^{J}
_{K}), and the initial values of **d**
_{k}, **b**
_{k}, **c**
_{k},*B _{k}*, where

*k*= 1,…,

*K*, as well as

*σ*

^{2}, we iterate the following iterative conditional sampling steps:

- for
*i*= 1,…,*n*, draw a new*j*from the conditional distribution from (14) to replace the old one;_{i} - conditional on
**J**, sequentially- update
**d**_{k}by a draw from [**d**_{k}|**b**_{k},**c**_{k},*σ*^{2},,**δ****y**^{J}_{k}], where*k*= 1,…,*K*, - update
**b**_{k}from [**b**_{k}|**d**_{k},**c**_{k},*σ*^{2},*B*,_{k}**y**^{J}_{k}], where*k*= 1,…,*K*, - update
**c**_{k}from [**c**_{k}|**d**_{k},**b**_{k},*σ*^{2},*τ*^{2}_{k},**y**^{J}_{k}], where*k*= 1,…,*K*, - update
*B*∼ [_{k}*B*|_{k}**b**_{k}], and*τ*^{2}_{k}∼ [*τ*^{2}_{k}|**c**_{k}], where*k*= 1,…,*K*, - update
*σ*^{2}∼ [*σ*^{2}|**d**,**b**,**c**,**y**], - update (
*p*_{1},…,*p*_{K}) ∼ Di(*n*_{1}+*α*_{1},…,*n*+_{K}*α*_{K}), where*n*is the number of genes in the_{j}*j*th cluster.

## 3. RESULTS AND DISCUSSION

To study oxygen-responsive gene network, Lai et al.
[11] used cDNA
microarray to monitor the gene expression changes of wild-type budding yeast (*Saccharomyces cerevisiae*) under aerobic condition in galactose medium.
Under aerobic condition, the oxygen concentration was lowered gradually until
oxygen was exhausted during a period of ten minutes. Microarray experiments
were conducted at 14 time points under aerobic condition. A reference sample
was obtained from a pooled RNA collected from all time points for
hybridization.

For the analysis, Lai et al. [11] normalized gene expression
after time 0 to gene expression of time 0 to set the common starting point.
They identified 2388 genes whose expression is differentially expressed at one
or more time points. Using our method, 2388 genes was clustered to 31 clusters,
which yields the smallest BIC. FunSpec [12] was used for gene annotation and biological function
enrichment analysis, where the Bonferroni-corrected functional enrichment *P*-values
based on hypergeometric distributions are reported.
We found 23 clusters out of 31 clusters discovered have biological functions
over-represented. Among them, estimated mean gene expression profiles of three
clusters are given in Figure 2.

In cluster A, which consists of 40 genes, the estimated
mean expression goes up progressively as oxygen level goes down, which suggests
that the genes in this cluster were transiently upregulated in response to
aerobisis. Accordingly, genes involved in stress response (function enrichment *P*-value = 10^{−4}) as well as
cell rescue and defense are over-represented in this cluster (function
enrichment *P*-value = 10^{−4}). Furthermore,
genes involved in molecular functions of oxidoreductase and coproporphyrinogen
oxidase are also presented, which explains the upregulation of the gene
expression levels.

We have 92 genes in cluster B, where the estimated
mean gene expression drops down at the beginning rapidly and then goes up
gradually. In this cluster, 34 genes are involved in protein synthesis
(function enrichment *P*-value ≤ 10^{−14}). Moreover,
ribosome biogenesis are also over-represented (function enrichment *P*-value ≤ 10^{−14}). These
processes were affected by oxygen level initially, but were quickly adjusted to
high expression levels to maintain living of yeast.

Contrast to cluster B, cluster C (68 genes) consists
of genes involved in galactose fermentation (function enrichment *P*-value = 10^{−4}), carbon
utilization (functional enrichment *P*-value = 10^{−2}), and
carbohydrate metabolism (function enrichment *P*-value ≤ 10^{−10}). The initial
upregulation of gene expression under aerobic condition can be partly explained
by the fact that the cell increases the energy
uptaking through the carbon utilization as oxygen level goes down; but as the
oxygen level continues to drop down, these processes are replaced by the more
energy-efficient processes, which drives the expression levels of genes to be
downregulated.

## 4. CONCLUSIONS

Conventional clustering methods do not take into consideration the correlation in the gene expression levels over time. Multivariate Gaussian models and time series analysis cannot model the time factor and correlation properly. These limitations can be readily overcome by the full Bayesian approach developed here. Although certain prior distributions and the related hyperparameters need to be input by the user, we found the clustering results rather robust to variations in such inputs. Moreover, our Bayesian clustering algorithm serves as a platform to incorporate more biological knowledge. Open source R code is available at www.stat.uiuc.edu/~pingma/BayesianFDAClust.htm.

## ACKNOWLEDGMENTS

The authors thank C. I. Castillo-Davis for his help in designing Figure 1 and for the many constructive suggestions and discussion in the early stage of this research. The authors thank Ji Young Kim for designing the software website. The authors also thank Kurt Kwast for providing the yeast microarray data.

## References

*Journal of the American Statistical Association*. 2003;98(462):397–408.

*Bioinformatics*. 2003;19(4):474–482. [PubMed]

*Bioinformatics*. 2004;20(3):332–339. [PubMed]

*Journal of the American Statistical Association*. 2006;101(473):18–29.

*Nucleic Acids Research*. 2006;34(4):1261–1269. [PMC free article] [PubMed]

*Annals of Statistics*. 2005;33(3):1357–1379.

*Eukaryotic Cell*. 2006;5(9):1468–1489. [PMC free article] [PubMed]

*BMC Bioinformatics*. 2002;3:3–35. [PMC free article] [PubMed]

**Hindawi Publishing Corporation**

## Formats:

- Article |
- PubReader |
- ePub (beta) |
- PDF (587K) |
- Citation

- Bayesian infinite mixture model based clustering of gene expression profiles.[Bioinformatics. 2002]
*Medvedovic M, Sivaganesan S.**Bioinformatics. 2002 Sep; 18(9):1194-206.* - Bayesian mixture model based clustering of replicated microarray data.[Bioinformatics. 2004]
*Medvedovic M, Yeung KY, Bumgarner RE.**Bioinformatics. 2004 May 22; 20(8):1222-32. Epub 2004 Feb 10.* - Analysis of a Gibbs sampler method for model-based clustering of gene expression data.[Bioinformatics. 2008]
*Joshi A, Van de Peer Y, Michoel T.**Bioinformatics. 2008 Jan 15; 24(2):176-83. Epub 2007 Nov 22.* - Clustering microarray gene expression data using weighted Chinese restaurant process.[Bioinformatics. 2006]
*Qin ZS.**Bioinformatics. 2006 Aug 15; 22(16):1988-97. Epub 2006 Jun 9.* - A dynamically growing self-organizing tree (DGSOT) for hierarchical clustering gene expression profiles.[Bioinformatics. 2004]
*Luo F, Khan L, Bastani F, Yen IL, Zhou J.**Bioinformatics. 2004 Nov 1; 20(16):2605-17. Epub 2004 May 6.*

- Inferring Dynamic Signatures of Microbes in Complex Host Ecosystems[PLoS Computational Biology. 2012]
*Gerber GK, Onderdonk AB, Bry L.**PLoS Computational Biology. 2012 Aug; 8(8)e1002624* - Information criterion-based clustering with order-restricted candidate profiles in short time-course microarray experiments[BMC Bioinformatics. ]
*Liu T, Lin N, Shi N, Zhang B.**BMC Bioinformatics. 10146*

- CompoundCompoundPubChem Compound links
- PubMedPubMedPubMed citations for these articles
- SubstanceSubstancePubChem Substance links
- TaxonomyTaxonomyRelated taxonomy entry
- Taxonomy TreeTaxonomy Tree

- Bayesian Functional Data Clustering for Temporal Microarray DataBayesian Functional Data Clustering for Temporal Microarray DataInternational Journal of Plant Genomics. 2008; 2008()

Your browsing activity is empty.

Activity recording is turned off.

See more...