Abstract

Biological measurements are a rich source of information about the biological phenomena that are represented. For example, time-series dynamic genomic or metabolic microarray data can be used to construct dynamic genetic regulatory network models, which can be used to better understand the interactions among different genes within the biological system and to design intervention strategies to cure or manage major diseases. Unfortunately, biological measurements are usually highly contaminated with errors that mask the important features in the data and limit their applicability. Therefore, these noisy measurements need to be filtered to enhance their usefulness in practice. In this work, various model-based and model-free data filtering techniques are used to denoise (or filter) genomic data. In the availability of a dynamic model representing the biological system, state estimation techniques, such as extended Kalman filtering (EKF), unscented Kalman filtering (UKF), and particle filtering (PF) are used to filter the measured data. When a model is not available, on the other hand, low-pass as well as multiscale filtering techniques will be utilized. Low-pass filters include the mean and exponentially weighted moving average filters, while the multiscale filters include several online as well as batch wavelet-based thresholding techniques. In this paper, the performances of all filtering techniques will be demonstrated and compared through their application using simulated time-series metabolic data contaminated with white noise. The results show clear advantages for the model-based over the model-free filtering techniques, and that the PF outperforms other model-based methods. The results also show that in the absence of a model of the biological system, the model-free filtering techniques, especially multiscale filtering, can also provide acceptable performances. Online multiscale (OLMS) filtering is shown to outperform low-pass filtering, and the batch multiscale methods, i.e., translation invariant (TI) and boundary corrected TI (BCTI) provide enhanced smoothness, with improved ability of BCTI over TI at the edges. From a biological perspective, the model-based and online model-free filtering techniques can be used when filtering is needed online, such as within an intervention framework to cure diseases, while the batch model-free filtering techniques can be used within a modeling framework to enhance the quality of the estimated biological models.

Original languageEnglish
Pages (from-to)109-121
Number of pages13
JournalNetwork Modeling and Analysis in Health Informatics and Bioinformatics
Volume2
Issue number3
DOIs
Publication statusPublished - 1 Jan 2013

Fingerprint

Biological Models
Biological Phenomena
Masks
Genes

Keywords

  • Filtering genomic data
  • Multiscale filtering
  • State estimation
  • Wavelets

ASJC Scopus subject areas

  • Urology

Cite this

@article{08ef635f24434f95a03dcf4fd69c2db2,
title = "Model-based and model-free filtering of genomic data",
abstract = "Biological measurements are a rich source of information about the biological phenomena that are represented. For example, time-series dynamic genomic or metabolic microarray data can be used to construct dynamic genetic regulatory network models, which can be used to better understand the interactions among different genes within the biological system and to design intervention strategies to cure or manage major diseases. Unfortunately, biological measurements are usually highly contaminated with errors that mask the important features in the data and limit their applicability. Therefore, these noisy measurements need to be filtered to enhance their usefulness in practice. In this work, various model-based and model-free data filtering techniques are used to denoise (or filter) genomic data. In the availability of a dynamic model representing the biological system, state estimation techniques, such as extended Kalman filtering (EKF), unscented Kalman filtering (UKF), and particle filtering (PF) are used to filter the measured data. When a model is not available, on the other hand, low-pass as well as multiscale filtering techniques will be utilized. Low-pass filters include the mean and exponentially weighted moving average filters, while the multiscale filters include several online as well as batch wavelet-based thresholding techniques. In this paper, the performances of all filtering techniques will be demonstrated and compared through their application using simulated time-series metabolic data contaminated with white noise. The results show clear advantages for the model-based over the model-free filtering techniques, and that the PF outperforms other model-based methods. The results also show that in the absence of a model of the biological system, the model-free filtering techniques, especially multiscale filtering, can also provide acceptable performances. Online multiscale (OLMS) filtering is shown to outperform low-pass filtering, and the batch multiscale methods, i.e., translation invariant (TI) and boundary corrected TI (BCTI) provide enhanced smoothness, with improved ability of BCTI over TI at the edges. From a biological perspective, the model-based and online model-free filtering techniques can be used when filtering is needed online, such as within an intervention framework to cure diseases, while the batch model-free filtering techniques can be used within a modeling framework to enhance the quality of the estimated biological models.",
keywords = "Filtering genomic data, Multiscale filtering, State estimation, Wavelets",
author = "Mohamed Nounou and Hazem Nounou and Majdi Mansouri",
year = "2013",
month = "1",
day = "1",
doi = "10.1007/s13721-013-0030-1",
language = "English",
volume = "2",
pages = "109--121",
journal = "Network Modeling and Analysis in Health Informatics and Bioinformatics",
issn = "2192-6662",
publisher = "Springer Nature",
number = "3",

}

TY - JOUR

T1 - Model-based and model-free filtering of genomic data

AU - Nounou, Mohamed

AU - Nounou, Hazem

AU - Mansouri, Majdi

PY - 2013/1/1

Y1 - 2013/1/1

N2 - Biological measurements are a rich source of information about the biological phenomena that are represented. For example, time-series dynamic genomic or metabolic microarray data can be used to construct dynamic genetic regulatory network models, which can be used to better understand the interactions among different genes within the biological system and to design intervention strategies to cure or manage major diseases. Unfortunately, biological measurements are usually highly contaminated with errors that mask the important features in the data and limit their applicability. Therefore, these noisy measurements need to be filtered to enhance their usefulness in practice. In this work, various model-based and model-free data filtering techniques are used to denoise (or filter) genomic data. In the availability of a dynamic model representing the biological system, state estimation techniques, such as extended Kalman filtering (EKF), unscented Kalman filtering (UKF), and particle filtering (PF) are used to filter the measured data. When a model is not available, on the other hand, low-pass as well as multiscale filtering techniques will be utilized. Low-pass filters include the mean and exponentially weighted moving average filters, while the multiscale filters include several online as well as batch wavelet-based thresholding techniques. In this paper, the performances of all filtering techniques will be demonstrated and compared through their application using simulated time-series metabolic data contaminated with white noise. The results show clear advantages for the model-based over the model-free filtering techniques, and that the PF outperforms other model-based methods. The results also show that in the absence of a model of the biological system, the model-free filtering techniques, especially multiscale filtering, can also provide acceptable performances. Online multiscale (OLMS) filtering is shown to outperform low-pass filtering, and the batch multiscale methods, i.e., translation invariant (TI) and boundary corrected TI (BCTI) provide enhanced smoothness, with improved ability of BCTI over TI at the edges. From a biological perspective, the model-based and online model-free filtering techniques can be used when filtering is needed online, such as within an intervention framework to cure diseases, while the batch model-free filtering techniques can be used within a modeling framework to enhance the quality of the estimated biological models.

AB - Biological measurements are a rich source of information about the biological phenomena that are represented. For example, time-series dynamic genomic or metabolic microarray data can be used to construct dynamic genetic regulatory network models, which can be used to better understand the interactions among different genes within the biological system and to design intervention strategies to cure or manage major diseases. Unfortunately, biological measurements are usually highly contaminated with errors that mask the important features in the data and limit their applicability. Therefore, these noisy measurements need to be filtered to enhance their usefulness in practice. In this work, various model-based and model-free data filtering techniques are used to denoise (or filter) genomic data. In the availability of a dynamic model representing the biological system, state estimation techniques, such as extended Kalman filtering (EKF), unscented Kalman filtering (UKF), and particle filtering (PF) are used to filter the measured data. When a model is not available, on the other hand, low-pass as well as multiscale filtering techniques will be utilized. Low-pass filters include the mean and exponentially weighted moving average filters, while the multiscale filters include several online as well as batch wavelet-based thresholding techniques. In this paper, the performances of all filtering techniques will be demonstrated and compared through their application using simulated time-series metabolic data contaminated with white noise. The results show clear advantages for the model-based over the model-free filtering techniques, and that the PF outperforms other model-based methods. The results also show that in the absence of a model of the biological system, the model-free filtering techniques, especially multiscale filtering, can also provide acceptable performances. Online multiscale (OLMS) filtering is shown to outperform low-pass filtering, and the batch multiscale methods, i.e., translation invariant (TI) and boundary corrected TI (BCTI) provide enhanced smoothness, with improved ability of BCTI over TI at the edges. From a biological perspective, the model-based and online model-free filtering techniques can be used when filtering is needed online, such as within an intervention framework to cure diseases, while the batch model-free filtering techniques can be used within a modeling framework to enhance the quality of the estimated biological models.

KW - Filtering genomic data

KW - Multiscale filtering

KW - State estimation

KW - Wavelets

UR - http://www.scopus.com/inward/record.url?scp=85049137532&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85049137532&partnerID=8YFLogxK

U2 - 10.1007/s13721-013-0030-1

DO - 10.1007/s13721-013-0030-1

M3 - Article

VL - 2

SP - 109

EP - 121

JO - Network Modeling and Analysis in Health Informatics and Bioinformatics

JF - Network Modeling and Analysis in Health Informatics and Bioinformatics

SN - 2192-6662

IS - 3

ER -