A Seminar about "Gene Expression Data Analysis Using Machine Learning Techniques" in FCITR
|
The advent of highly parallel genomic platforms like microarray
technology has facilitated a principal transition from gene science to
genome science. Microarray provides the researchers with enormous data
for the expression levels of thousands of genes measured at once. Full
exploitation of microarray data aims to explore the complex
relationships between genes and other regulatory network components that
underpin all biological processes. However, microarray data contains
little information about how these genes are regulated, and needs
sophisticated machine learning methods to achieve this task.
Quantitative estimation of the regulatory relationship between
transcription factors and genes is a key problem when trying to model
the gene regulatory network. Because the difficulty of measuring the
concentration levels of transcription factors and the fact that
transcription factors are post-transcriptionally regulated, most of the
work in the literature has been dedicated to infer the regulatory
relationship between the transcription factors and the target genes from
the expression levels of these genes.
In this work, a novel probabilistic model based on Gaussian Mixture
Regression (GMR) is presented to derive the relationship between the
transcription factor protein and each of its target genes quantitatively
by estimating the sensitivity of these genes to the transcription
factor’s activity; and to generate a ranked list of predicted genes.
The potential power of GMR lies in its structure that combines the
advantages of parametric models and the flexibility of non-parametric
models. The GMR model is trained using microarray time series data and
the verification scores, which are inferred using small interfering RNA.
The model is applied to the p53 network and the results are compared
with similar work based on differential equations. The investigated
model provides us with a practical toolkit that can be applied to other
transcription factors. The obtained information can be further exploited
to build intelligent cancer classification systems.
|