Modelling-based experiment retrieval: a case study with gene expression clustering.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Public and private experimental data repositories are rapidly expanding.
• Current keyword search methods for experimental data are limited.
• Content-based retrieval methods for gene expression data are an area of active research.

Purpose of the Study:

• To develop an improved method for retrieving relevant gene expression experiments.
• To address limitations of existing retrieval methods, particularly for noisy and high-dimensional data.
• To propose a general probabilistic framework for data retrieval.

Main Methods:

• Developed a general probabilistic framework for modeling individual experiments.
• Utilized a product partition model to cluster genes with similar expression patterns.
• Employed normalized information distance as a metric for retrieval.
• Approximated full probabilistic model inference using heuristic clustering (e.g., k-means).

Main Results:

• Retrieval using denoised models significantly outperforms retrieval using original noisy data.
• The proposed method is highly scalable and straightforward to implement.
• Heuristic clustering provides good performance approximations for the probabilistic model.
• Empirical results demonstrate the effectiveness of the denoised retrieval approach.

Conclusions:

• The proposed method offers a scalable and effective solution for gene expression experiment retrieval.
• Denoising query datasets enhances the accuracy of content-based retrieval.
• The framework is adaptable for constructing general-purpose data retrieval systems.