Recursive SVM Feature Selection and Sample Classification for Mass-Spectrometry and Microarray Data
Author Information
Author(s): Zhang Xuegong, Lu Xin, Shi Qian, Xu Xiu-qin, Leung Hon-chiu E, Harris Lyndsay N, Iglehart James D, Miron Alexander, Liu Jun S, Wong Wing H
Primary Institution: Tsinghua University
Hypothesis
Can the recursive support vector machine (R-SVM) algorithm improve the selection of informative genes and classification of noisy data compared to existing methods?
Conclusion
The R-SVM method is effective for analyzing noisy high-throughput proteomics and microarray data, outperforming SVM-RFE in robustness and feature recovery.
Supporting Evidence
- The R-SVM method showed a 5%-20% improvement over SVM-RFE in recovering informative genes.
- R-SVM was more robust to noise and outliers compared to SVM-RFE.
- Important biomarkers identified by R-SVM were validated through follow-up biological experiments.
Takeaway
This study created a new method to help find important genes in noisy data, which can help doctors better understand diseases like cancer.
Methodology
The study developed a recursive support vector machine (R-SVM) algorithm and compared its performance with SVM-RFE and univariate methods on simulated and real proteomics datasets.
Potential Biases
The feature selection procedure should be validated simultaneously with the classification to avoid bias.
Limitations
The study's simulation models may favor univariate methods, and the performance may vary with different settings of parameters.
Participant Demographics
The study involved 75 breast cancer patients and 61 healthy women for the human breast cancer dataset, and 26 normal rats and 69 cirrhosis rats for the rat liver cirrhosis dataset.
Statistical Information
P-Value
p<0.01
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website