Positive-unlabeled learning for disease gene identification

Yang, Peng; Li, Xiaoli; Mei, Jian-Ping; Kwoh, Chee-Keong; Ng, See-Kiong

doi:10.1093/bioinformatics/bts504

Cited by 173 publications

(135 citation statements)

References 36 publications

Supporting

Mentioning

133

Contrasting

Order By: Relevance

“…Therefore, to reduce this uncertainty, the binary-class supervised learning methods have to choose and were different in the way to construct the set of negative training samples. In reality, the unknown set may contain unknown disease genes; therefore, semi-supervised learning methods were recently proposed to solve the problem, where the classifier is learned from both labeled (i.e., known disease genes) and unlabeled (i.e., the unknown genes) data [50,51]. Another recent approach to disease gene prediction is to use unary/one-class classification techniques, in which the classifier is learned from only positive samples (i.e., known disease genes) [52].…”

Section: Discussionmentioning

confidence: 99%

A Comparative Study of Classification-Based Machine Learning Methods for Novel Disease Gene Prediction

Hoài²,

Kwon

2015

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

Abstract. Prediction of novel genes associated to a disease is an important issue in biomedical research. At early days, annotation-based methods were proposed for this problem. In next stage, with high-throughput technologies, data of interaction between genes/proteins has grown quickly and covered almost genome and proteome, and therefore network-based methods for the issue is becoming prominent. Besides those two methods, the prediction problem can be also approached using machine learning techniques because it can be formulated as a classification task of machine learning. To date, a number of supervised learning techniques and various types of gene/protein annotation data have been used to solve the disease gene classification/prediction problem. However, to the best of our knowledge, there has been no study on the comparison of these methods that work on comprehensive biomedical annotation data. In addition, it is generally true that no classifier is better than others for all classification problems. Therefore, in this study, we compare the performance of disease gene prediction of several supervised learning techniques that have been used in the literature such as Decision Tree Learning, k-Nearest Neighbor, Naive Bayesian, Artificial Neural Networks and Support Vector Machines. We additionally assess Random Forest, a relatively new decision-tree-based ensemble learning method. The simulation results indicate that Random Forest obtained the best performance of all. Also, all methods are stable with the change of known disease genes used as positive training samples.

show abstract

Section: Discussionmentioning

confidence: 99%

A Comparative Study of Classification-Based Machine Learning Methods for Novel Disease Gene Prediction

Hoài²,

Kwon

2015

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

show abstract

“…In our experiments, we have employed employed the data used by (Yang et al, 2012). This data has 5405 known disease genes spanning 2751 disease phenotypes, where all the genes have been extracted by combining GENECARD (Safran et al, 2010) and OMIM (McKusick, 2007) disease gene data.…”

Section: Experimental Datamentioning

confidence: 99%

“…In this regard, the performance of the final classifier will be decreased. Yang et al (2012) (PUDI), selected some samples from unknown genes by applying the Euclidean distance between 'positive representative vector' and each of the unknown genes as negative instances. They also defined three other sets namely, likely negative, likely positive, and weakly negative set, based on their likelihoods to be positive or negative class.…”

Section: Introductionmentioning

confidence: 99%

“…In the first layer, four sequencetranslated methods, which are based on physicochemical properties of amino acid, are employed to construct four different feature vectors for each gene products (proteins). Since there is no information available about negative data (non-disease genes), we have selected some reliable negative data from unknown genes by using a PUDI method (Yang, et al, 2012) with a cosine similarity instead of Euclidean distance in the second layer. Then, each feature vector is trained using SVM algorithm separately, in the third layer.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

SFM: A novel sequence-based fusion method for disease genes identification and prioritization

Yousef

Charkari

2015

Journal of Theoretical Biology

View full text Add to dashboard Cite

“…One common approach is Positive and Unlabeled (PU) learning [83] that learns from positive and unlabeled data alone. It is used when only the labels for disease genes are available [130,131]. the maximum class probability greater than 0.5; otherwise the unknown class would be predicted.…”

Section: Positive and Unlabeled Learning Approachesmentioning

confidence: 99%

Healthcare data mining from multi-source data

Chen¹

View full text Add to dashboard Cite

The "big data" challenge is changing the way we acquire, store, analyse, and draw conclusions from data. How we effectively and efficiently "mine" the data from possibly multiple sources and extract useful information is a critical question. Increasing research attention has been drawn to healthcare data mining, with an ultimate goal to improve the quality of care. The human body is complex and so too the data collected in treating it. Data noise that is often introduced via the collection process makes building Data Mining models a challenging task.This thesis focuses on the classification tasks of mining healthcare data, with the goal of improving the effectiveness of health risk prediction. In particular, we developed algorithms to address issues identified from real healthcare data, such as feature extraction, heterogeneity, label uncertainty, and large unlabeled data.The three main contributions of this research are as follows. First, we developed a new health index called Personal Health Index (PHI) that scores a person's health status based on the examination records of a given population. Second, we identified the key characteristics of the real datasets and issues that were associated with the data. Third, we developed classification algorithms to cope with those issues, particularly, the label uncertainty and large unlabeled data issues.This research takes one step forward towards scoring personal health based on mining increasingly large health records. Particularly, it pioneers exploring the mining of GHE data and tackles the associated challenges. It is our anticipation that in the near future, more robust data-mining-based health scoring systems will be available for healthcare professionals to understand people's health status and thus improve the quality of care.

show abstract

Positive-unlabeled learning for disease gene identification

Cited by 173 publications

References 36 publications

A Comparative Study of Classification-Based Machine Learning Methods for Novel Disease Gene Prediction

A Comparative Study of Classification-Based Machine Learning Methods for Novel Disease Gene Prediction

SFM: A novel sequence-based fusion method for disease genes identification and prioritization

Healthcare data mining from multi-source data

Contact Info

Product

Resources

About