[Paper Review] Gene selection for cancer classification using a hybrid of univariate and multivariate feature selection methods

Min Xu, Rudy Setiono|arXiv (Cornell University)|Jun 5, 2015

Gene expression and cancer classification22 references5 citations

TL;DR

This paper proposes a hybrid gene selection method combining univariate Maximum Likelihood (LIK) and multivariate Recursive Feature Elimination (RFE) to improve cancer classification accuracy with fewer genes. By leveraging LIK for initial gene ranking and RFE for iterative feature elimination, the method reduces noise sensitivity and computational cost while achieving superior or equal accuracy with smaller gene sets on leukemia and SRBCT datasets.

ABSTRACT

Various approaches to gene selection for cancer classification based on microarray data can be found in the literature and they may be grouped into two categories: univariate methods and multivariate methods. Univariate methods look at each gene in the data in isolation from others. They measure the contribution of a particular gene to the classification without considering the presence of the other genes. In contrast, multivariate methods measure the relative contribution of a gene to the classification by taking the other genes in the data into consideration. Multivariate methods select fewer genes in general. However, the selection process of multivariate methods may be sensitive to the presence of irrelevant genes, noises in the expression and outliers in the training data. At the same time, the computational cost of multivariate methods is high. To overcome the disadvantages of the two types of approaches, we propose a hybrid method to obtain gene sets that are small and highly discriminative. We devise our hybrid method from the univariate Maximum Likelihood method (LIK) and the multivariate Recursive Feature Elimination method (RFE). We analyze the properties of these methods and systematically test the effectiveness of our proposed method on two cancer microarray datasets. Our experiments on a leukemia dataset and a small, round blue cell tumors dataset demonstrate the effectiveness of our hybrid method. It is able to discover sets consisting of fewer genes than those reported in the literature and at the same time achieve the same or better prediction accuracy.

Motivation & Objective

• To address the limitations of univariate and multivariate gene selection methods in cancer classification.
• To reduce sensitivity to irrelevant genes, noise, and outliers in multivariate methods.
• To lower the high computational cost of multivariate approaches.
• To identify smaller, highly discriminative gene sets that maintain or improve classification accuracy.
• To integrate univariate and multivariate methods into a synergistic hybrid framework.

Proposed method

• The method combines univariate Maximum Likelihood (LIK) for initial gene ranking based on individual gene contribution to classification.
• It applies multivariate Recursive Feature Elimination (RFE) to iteratively remove less relevant genes while considering gene interactions.
• Gene selection begins with LIK to rank genes, followed by RFE to refine the set based on multivariate relevance.
• The hybrid approach reduces the impact of noise and irrelevant features by filtering them early via LIK before RFE's computationally intensive process.
• The method is evaluated on two microarray datasets: acute leukemia and small round blue cell tumors (SRBCT).
• Feature selection is validated using classification accuracy as the primary performance metric.

Experimental results

Research questions

• RQ1 Can a hybrid approach combining univariate and multivariate methods improve gene selection for cancer classification?
• RQ2 Does integrating LIK with RFE reduce the number of selected genes while maintaining or improving classification accuracy?
• RQ3 How does the proposed method compare to existing univariate and multivariate methods in terms of robustness to noise and outliers?
• RQ4 To what extent does the hybrid method reduce computational cost compared to pure multivariate methods?
• RQ5 Can the hybrid method identify smaller, more discriminative gene sets than those reported in the literature?

Key findings

• The hybrid method achieved classification accuracy comparable to or better than state-of-the-art methods on both the leukemia and SRBCT datasets.
• The method selected significantly fewer genes than previously reported results while maintaining high predictive performance.
• On the leukemia dataset, the method identified a small gene set that matched or exceeded the accuracy of larger gene sets in prior studies.
• On the SRBCT dataset, the hybrid approach outperformed individual univariate and multivariate methods in terms of gene set size and accuracy.
• The integration of LIK and RFE improved robustness to noise and outliers compared to standalone multivariate methods.
• The computational cost was reduced due to the pre-filtering step using LIK, which limited the number of genes passed to the more expensive RFE phase.