Syng-Yup Ohn

Feature Selection for Classification of Mass Spectrometric Proteomic Data Using Random Forest

ABSTRACTThis paper proposes a novel method for feature selection for mass spectrometric proteomic data based on Random Forest. The method includes an effective preprocessing step to filter a large amount of redundant features with high correlation and applies a tournament strategy to get an optimal feature subset. Experiments on three public datasets, Ovarian 4-3-02, Ovarian 7-8-02 and Prostate shows that the new method achieves high performance comparing with widely used methods and balanced rate of specificity and sensitivity. Key words : Feature Selection, Bioinformatics, Pattern Recognition, SELDI-TOF, Proteome, Spectrum, Random Forest, Pearson Correlation요 약본 논문에서는 질량 분석 방법에 의하여 산출된 단백체 데이터 (mass spectrometric proteomic data)의 분류 분석(classification analysis) 을 위한 새로운 특성 선택 (feature selection) 방법을 제안한다. 이 방법은 i)높은 상관관계를 가지는 중복된 특성을 효과적으로 제거하는 전처리 단계와 ii)토너먼트(tournament) 전략을 사용하여 최적 특성 부분집합 (optimal feature subset)을 탐색해 내는 단계로 구성되어 있다 . 제안 되는 방법을 실제 암진단에 사용되는 공개된 혈액 단백체 데이터에 적용하였으며 널리 사용되는 타 방법과 비교할 때 우수한 성능과 균형된 특이도와 민감도를 달성함을 실증하였다 .주요어 : 특성 선택, 생물정보학, 패턴인식, 표면 강화한 레이저 탈착과 이온화 시간 의 비행 질량 분석 , 단백체, 스펙트럼, 랜덤 포리스트, 피어슨 상관 계수

RISC: a new filter approach for feature selection from proteomic data

This paper proposes a novel feature selection technique for SELDITOF spectrum data. The new technique, called RISC (Relevance Index by Sample Counting), measures the relevance of features based on each samples discriminating power to partition the samples in the opposite class. We also proposes a heuristic searching method to obtain the optimal feature set, which makes use of the relevance parameters. Our technique is fast even for extremely high-dimensional datasets such as SELDI spectrum, since it has low computational complexity and consists of simple counting operations. The new technique also shows good performance comparable to the conventional feature selection techniques from the experiment on three clinical datasets from NCI/CCR and FDA/CBER Clinical Proteomics Program Databank: Ovarian 4-3-02, Ovarian 7-8-02, Prostate.

Feature elimination approach based on random forest for cancer diagnosis

The performance of learning tasks is very sensitive to the characteristics of training data. There are several ways to increase the effect of learning performance including standardization, normalization, signal enhancement, linear or non-linear space embedding methods, etc. Among those methods, determining the relevant and informative features is one of the key steps in the data analysis process that helps to improve the performance, reduce the generation of data, and understand the characteristics of data. Researchers have developed the various methods to extract the set of relevant features but no one method prevails. Random Forest, which is an ensemble classifier based on the set of tree classifiers, turns out good classification performance. Taking advantage of Random Forest and using wrapper approach first introduced by Kohavi et al, we propose a new algorithm to find the optimal subset of features. The Random Forest is used to obtain the feature ranking values. And these values are applied to decide which features are eliminated in the each iteration of the algorithm. We conducted experiments with two public datasets: colon cancer and leukemia cancer. The experimental results of the real world data showed that the proposed method results in a higher prediction rate than a baseline method for certain data sets and also shows comparable and sometimes better performance than the feature selection methods widely used.

Cancer Diagnosis System using Genetic Algorithm and Multi-boosting Classifier

It is believed that the anomalies or diseases of human organs are identified by the analysis of the patterns. This paper proposes a new classification technique for the identification of cancer disease using the proteome patterns obtained from two-dimensional polyacrylamide gel electrophoresis(2-D PAGE). In the new classification method, three different classification methods such as support vector machine(SVM), multi-layer perceptron(MLP) and k-nearest neighbor(k-NN) are extended by multi-boosting method in an array of subclassifiers and the results of each subclassifier are merged by ensemble method. Genetic algorithm was applied to obtain optimal feature set in each subclassifier. We applied our method to empirical data set from cancer research and the method showed the better accuracy and more stable performance than single classifier.

Unified kernel function and its training method for SVM

This paper proposes a unified kernel function for support vector machine and its learning method with a fast convergence and a good classification performance. We defined the unified kernel function as the weighted sum of a set of different types of basis kernel functions such as neural, radial, and polynomial kernels, which are trained by a new learning method based on genetic algorithm. The weights of basis kernel functions in the unified kernel are determined in learning phase and used as the parameters in the decision model in the classification phase. The unified kernel and the learning method were applied to obtain the optimal decision model for the classification of two public data sets for diagnosis of cancer diseases. The experiment showed fast convergence in learning phase and resulted in the optimal decision model with the better performance than other kernels. Therefore, the proposed kernel function has the greater flexibility in representing a problem space than other kernel functions.