2 단계 접근법을 통한 통합 마이크로어레이 데이타의 분류기 생성

윤영미(Youngmi Yoon),이종찬(Jongchan Lee),박상현(Sanghyun Park) 한국정보과학회 2007 정보과학회논문지 : 데이타베이스 Vol.34 No.1

마이크로어레이 데이타는 동시에 수 만개 유전자의 발현 값을 포함하고 있기 때문에 질병의 발현 형질 분류에 매우 유용하게 쓰인다. 그러나 동일한 생물학적 주제라 할지라도 여러 독립된 연구 집단에서 생성된 마이크로어레이의 분석결과는 서로 다르게 나타날 수 있다. 이에 대한 주된 이유는 하나의 마이크로어레이 실험에 참여한 샘플의 수가 제한적이기 때문이다. 따라서 개별적으로 수행된 마이크로어레이 데이타를 통합하여 샘플의 수를 늘리는 것은, 보다 정확한 분석을 하는데 있어 매우 중요하다. 본 연구에서는 이에 대한 해결 방안으로 두 단계 접근방법을 제안한다. 제 1 단계에서는 개별적으로 생성된 동일 주제의 마이크로어레이 데이타를 통합한 후 인포머티브(Informative) 유전자를 추출하고 제 2 단계에서는 인포머티브 유전자만을 이용하여 클래스 분류(Classification) 과정 후 분류자를 추출한다. 이 분류자를 다른 테스트 샘플 데이타에 적용한 실험결과를 보면 마이크로어레이 데이타를 통합하여 샘플의 수를 증가시킬수록, 비교 방법에 비해 정확도가 최대 24.19% 높은 분류자를 만들어 내는 것을 알 수 있다. Since microarray data acquire tens of thousands of gene expression values simultaneously, they could be very useful in identifying the phenotypes of diseases. However, the results of analyzing several microarray datasets which were independently carried out with the same biological objectives, could turn out to be different. One of the main reasons is attributable to the limited number of samples involved in one microarry experiment. In order to increase the classification accuracy, it is desirable to augment the sample size by integrating and maximizing the use of independently-conducted microarray datasets. In this paper, we propose a novel two-stage approach which firstly integrates individual microarray datasets to overcome the problem caused by limited number of samples, and identifies informative genes, secondly builds a classifier using only the informative genes. The classifier from large samples by integrating independent microarray datasets achieves high accuracy up to 24.19% increase as against other comparison methods, sensitivity, and specificity on independent test sample dataset.

협력적 필터링 기술에서 평균 정보량인 엔트로피를 이용한 효율적인 예측 방법

윤영미(YoungMi Yoon),정경용(KyungYong Jung) 한국정보기술학회 2003 한국정보기술학회논문지 Vol.1 No.1

Existing internet web site is applying Collaborative filtering way to provide service that is personalized by user to maximize user's satisfaction. Collaborative filtering technology forecasts and recommends an appropriate item in accordance with user's inclination, and uses Pearson Correlation Coefficient usually to find correlation with other users that have similar preference. But, method to use Pearson Correlation Coefficient has shortcomings that it can find correlation when there is item that user evaluates only, and that the accuracy of prediction is low. Therefore, this paper proposes the method to measure accurate user's similarity, in order to complement the technique of prediction based on Pearson Correlation relation Proposed method applied entropy to preference degree of item that user votes, and implemented more accurate Collaborative filtering method by applying Default voting method about item that user does not vote preference degree. As a result, we have proven that the proposed method in this paper is effective to measure accuracy user similarity.

판별분석을 이용한 청소년 자살 잠재성 예측 모델

김정태(Jungtae Kim),윤영미(Youngmi Yoon) 한국정보기술학회 2015 한국정보기술학회논문지 Vol.13 No.5

Out of the many reasons of adolescents’ death, committing suicide takes the most part in South Korean Society. According to the cause of death statistics from the National Statistical Office(NSO), steadily rising from 1993 to 2011, and South Korea has shown the most suicide death rate out of all OECD countries. In Korea, various studies are at practice to find out the reason for this tragedy, but research on predicting suicide risk directly is still incomplete. Accordingly we need a model that can predict whether the risk of suicide in young people stays. In our study, we developed a model to predict suicide potential of Korean youth by applying the discriminant analysis based on data mining to the data set of KYRBS(The Ninth Korea Youth Risk Behavior Web-based Survey). The prediction model we developed showed a high accuracy of 78.1%. We are expecting that this will help preventing Korean adolescent from committing suicide.

텍스트 마이닝을 활용한 바이오 네트워크 구축의 동향

김현진(Hyunjin Kim),윤영미(Youngmi Yoon) 한국정보과학회 2015 정보과학회지 Vol.33 No.4

마이크로어레이 데이터와 임상인자의 통합데이터를 이용한 전립선암의 예후 예측

서보연(Boyeon Seo),윤영미(Youngmi Yoon) 한국정보기술학회 2012 한국정보기술학회논문지 Vol.10 No.5

Prognosis implies a medical prediction for the progress and development of disease. Prognosis prediction can dramatically improve the survival rate and quality of life since appropriate personalized treatment is possible. Recently for a prognosis prediction, microarry data which is gene expression information, has been used. This paper increased the classification accuracy of prognosis prediction by integrating gene expression information and clinical factors from clinical diagnosis. The experiments of this paper used microarray data of prostate cancer patients and clinical information data. This study compared and analyzed the results of 4 cases: gene expression data only, feature selected gene expression data, clinical factors only, and integrated data of 2nd and 3rd cases. In conclusion we were able to secure the highest prognosis prediction accuracy rate in the case of integrated data of gene expression data and clinical factors.

밈(Meme) 데이터의 패턴 분석 및 예측 시스템

주지훈(Jihoon Joo),윤영미(Youngmi Yoon) 한국정보기술학회 2011 한국정보기술학회논문지 Vol.9 No.9

We made an analysis on the pattern of on-line news which change as time flows. The record for the number of mentions in a series of time for short units of text that propagate and diffuse over the web from mainstream media to blogs is referred as “meme” data. The flow of changes in the number of quotes of these meme data is clustered using EM, K-means, X-means algorithm, and their patterns are compared. We also propose a prediction system for classifying the pattern of an independent meme with a set of training data using Naive Bayes, SMO, Random Forest. The training data set consists of the meme data where only the data prior to the peak are used, and the class label from the clustering. We validated the prediction system with Precision, Recall, F_measure, Accuracy using LOOCV. The accuracy rate is 95.3% when SMO classification algorithm is executed with the training data from K-means clustering algorithm. Our system will successfully predict the pattern of the independent meme data posterior to the peak.

PPI 네트워크상의 거리기반 속성을 이용한 드러그-리포지셔닝

오민(Min Oh),윤영미(Youngmi Yoon) 한국정보기술학회 2013 한국정보기술학회논문지 Vol.11 No.12

Finding a new purpose for FDA approved drugs are called drug-repositioning. The growing number of various omics datasets makes it more feasible to understand how drugs and diseases are associated on the molecular level. Proper features from the network representation of exiting drug-disease associations can be used to infer novel indications of existing drugs, with increased accuracy, and more concrete evidence. To find new drug-disease associations, we used protein-protein interaction. Similarities in drug-drug, disease-disease, or drug-disease of existing drug-disease associations were quantified and used as features for the prediction of novel drug-disease associations. The classification AUC of cross-validation was at least 0.938. The newly identified 21,180 drug-disease pairs have potential to be commercial drugs once further biological experiments are done by pharmacologists.

향상된 FFP(Feature Frequency Profile)을 활용한 악성 댓글의 판별시스템

김현정(HyunJung Kim),윤영미(YoungMi Yoon),이병문(ByungMun Lee) 한국정보기술학회 2011 한국정보기술학회논문지 Vol.9 No.1

Putting postings on other people's articles gives good means of communication. However, it is growing to use them in order to intrude privacy, make a personal attack or defamation behind a veil of anonymity. We propose a prediction system for abusive postings. FFP(Feature Frequency Profile) method is used for extracting features which exhibit frequently in the abusive posting. However FFP method has lack of interpreting different features of which meaning are same into same features since language use on internet often does not follow its formal rules. Our algorithm uses Unicode in order to handle linguistically destructed different features of which meanings are same. This method enhances FFP by replacing frequently exhibited words with initial consonants only for feature selection, and uses SVM and Random Forest for classification. This Enhanced FFP method achieves high accuracy up to 15.4% increase as against other comparison methods.

약물 리포지셔닝 분야에서 단백질 위치 정보의 활용 가능성에 대한 연구

여윤구(Yunku Yeu),윤영미(Youngmi Yoon),박상현(Sanghyun Park) 한국정보과학회 2014 한국정보과학회 학술발표논문집 Vol.2014 No.6

약물-질병 이분 네트워크를 통한 약물 재창출

유해강(Haekang Yu),윤영미(Youngmi Yoon) 한국정보기술학회 2020 한국정보기술학회논문지 Vol.18 No.12

The rate of developing new drugs has decreased significantly over the past few years. Drug repositioning is drawing attention as a way to solve this problem. Drug repositioning can solve the time and cost of drug development by finding new indications for previously proven drugs. In this study, drug-related data and disease-related data are used to measure drug-to-disease similarities, and to establish a drug-to-disease bifurcation network. New drug-disease relationships are sought through drugs with common indications in bipartite networks. Measuring AUC(Area Under the ROC Curve) values, this study method showed better results than GBA (Guilt By Association). Fischers exact test confirmed that the predicted candidate drug-disease relationship in this study was statistically significant in the KEGG database.