Jae-Hong Eom

AptaCDSS-E: A classifier ensemble-based clinical decision support system for cardiovascular disease level prediction

Conventional clinical decision support systems are generally based on a single classifier or a simple combination of these models, showing moderate performance. In this paper, we propose a classifier ensemble-based method for supporting the diagnosis of cardiovascular disease (CVD) based on aptamer chips. This AptaCDSS-E system overcomes conventional performance limitations by utilizing ensembles of different classifiers. Recent surveys show that CVD is one of the leading causes of death and that significant life savings can be achieved if precise diagnosis can be made. For CVD diagnosis, our system combines a set of four different classifiers with ensembles. Support vector machines and neural networks are adopted as base classifiers. Decision trees and Bayesian networks are also adopted to augment the system. Four aptamer-based biochip data sets including CVD data containing 66 samples were used to train and test the system. Three other supplementary data sets are used to alleviate data insufficiency. We investigated the effectiveness of the ensemble-based system with several different aggregation approaches by comparing the results with single classifier-based models. The prediction performance of the AptaCDSS-E system was assessed with a cross-validation test. The experimental results show that our system achieves high diagnosis accuracy (>94%) and comparably small prediction difference intervals (<6%), proving its usefulness in the clinical decision process of disease diagnosis. Additionally, 10 possible biomarkers are found for further investigation.

Evolutionary hypernetwork models for aptamer-based cardiovascular disease diagnosis

We present a biology-inspired probabilistic graphical model, called the hypernetwork model, and its application to medical diagnosis of disease. The hypernetwork models are a way of simulated DNA computing. They have a set of hyperedges representing a subset of features in the training data. These characteristics allow the hypernetwork models to work similarly to associative memories and make their learning results more understandable. This comprehensibility is one of main advantages of the models over other machine learning algorithms such as support vector machines and artificial neural networks which are used in a wide range of applications but are not easy to understand their learning results. Since medical applications require both competitive performance and understandability of results, the hypernetwork models are suitable for this kind of applications. However, ordinary hypernetwork models have limitations that hyperedges cannot be changed after they are sampled once. To improve this diversity problem, we adopted simple evolutionary computation method, the hyperedges replacement strategy as the method of keeping the diversity into conventional hypernetworks in addition to error correction for model learning. To show the improvement, we used aptamer-based cardiovascular disease data. Experiment results show that the hypernetworks can achieve fairly competitive performance and the results are also comprehensible.

PubMiner: Machine Learning-Based Text Mining System for Biomedical Information Mining

PubMiner, an intelligent machine learning based text mining system for mining biological information from the literature is introduced. PubMiner utilize natural language processing and machine learning based data mining techniques for mining useful biological information such as protein-protein interaction from the massive literature data. The system recognizes biological terms such as gene, protein, and enzymes and extracts their interactions described in the document through natural language analysis. The extracted interactions are further analyzed with a set of features of each entity which were constructed from the related public databases to infer more interactions from the original interactions. An inferred interaction from the interaction analysis and native interaction are provided to the user with the link of literature sources. The evaluation of system performance proceeded with the protein interaction data of S.cerevisiae (bakers yeast) from MIPS and SGD.

A tree kernel-based method for protein-protein interaction mining from biomedical literature

As genomic research advances, the knowledge discovery from a large collection of scientific papers becomes more important for efficient biological and biomedical research. Even though current databases continue to update new protein-protein interactions, valuable information still remains in biomedical literature. Thus data mining techniques are required to extract the information. In this paper, we present a tree kernel-based method to mine protein-protein interactions from biomedical literature. The tree kernel is designed to consider grammatical structures for given sentences. A support vector machine classifier is combined with the tree kernel and trained on predefined interaction corpus and set of interaction patterns. Experimental results show that the proposed method gives promising results by utilizing the structure patterns.

Prediction of Implicit Protein-Protein Interaction by Optimal Associative Feature Mining

Proteins are known to perform a biological function by interacting with other proteins or compounds. Since protein-protein interaction is intrinsic to most cellular processes, protein interaction prediction is an important issue in post-genomic biology where abundant interaction data has been produced by many research groups. In this paper, we present an associative feature mining method to predict implicit protein-protein interactions of S.cerevisiae from public protein-protein interaction data. To overcome the dimensionality problem of conventional data mining approach, we employ feature dimension reduction filter (FDRF) method based on the information theory to select optimal informative features and to speed up the overall mining procedure. As a mining method to predict interaction, we use association rule discovery algorithm for associative feature and rule mining. Using the discovered associative feature we predict implicit protein interactions which have not been observed in training data. According to the experimental results, the proposed method accomplishes about 94.8% prediction accuracy with reduced computation time which is 32.5% faster than conventional method that has no feature filter.

Adaptive neural network-based clustering of yeast protein: protein interactions

In this paper, we presents an adaptive neural network based clustering method to group protein–protein interaction data according to their functional categories for new protein interaction prediction in conjunction with information theory based feature selection. Our technique for grouping protein interaction is based on ART-1 neural network. The cluster prototype constructed with existing protein interaction data is used to predict the class of new protein interactions. The protein interaction data of S.cerevisiae (bakers yeast) from MIPS and SGD are used. The clustering performance was compared with traditional k-means clustering method in terms of cluster distance. According to the experimental results, the proposed method shows about 89.7% clustering accuracy and the feature selection filter boosted overall performances about 14.8%. Also, inter-cluster distances of cluster constructed with ART-1 based clustering method have shown high cluster quality.

Neural feature association rule mining for protein interaction prediction

The prediction of protein interactions is an important problem in post–genomic biology. In this paper, we present an association rule mining method for protein interaction prediction. A neural network is used to cluster protein interaction data and a feature selection is used to reduce the dimension of protein features. For model training, the preliminary network model was constructed with existing protein interaction data in terms of their functional categories and interactions. A set of association rules for protein interaction prediction are derived by decoding a set of learned weights of trained neural network after this model training. The protein interaction data of Yeast from public databases are used. The prediction performance was compared with simple association rule-based approach. According to the experimental results, proposed method achieved about 95.5% accuracy.

Mining protein interaction from biomedical literature with relation kernel method

Many interaction data still exist only in the biomedical literature and they require much effort to construct well-structured data. Discovering useful knowledge from large collections of papers is becoming more important for efficient biological and biomedical researches as genomic research advances. In this paper, we present a relation kernel-based interaction extraction method to extract knowledge efficiently. We extract protein interactions of from text documents with relation kernel and Yeast was used as an example target organism. Kernel for relation extraction is constructed with predefined interaction corpus and set of interaction patterns. The proposed method only exploits shallow parsed documents. Experimental results show that the proposed kernel method achieves a recall rate of 79.0% and precision rate of 80.8% for protein interaction extraction from biomedical document without full parsing efforts.

Prediction of the human papillomavirus risk types using gap-spectrum kernels

Human Papillomavirus (HPV) is known as the main cause of cervical cancer and classified to low- or high-risk type by its malignant potential. Detection of high-risk HPVs is critical to understand the mechanisms and recognize potential patients in medical judgments. In this paper, we present a simple kernel approach to classify HPV risk types from E6 protein sequences. Our method uses support vector machines combined with gap-spectrum kernels. The gap-spectrum kernel is introduced to compute the similarity between amino acids pairs with a fixed distance, which can be useful for the helical structure of proteins. In the experiments, the proposed method is compared with a mismatch kernel approach in accuracy and F1-score, and the predictions for unknown types are presented.

Extraction of gene/protein interaction from text documents with relation kernel

Even though there are many databases for gene/protein interactions, most such data still exist only in the biomedical literature. They are spread in biomedical literature written in natural languages and they require much effort such as data mining for constructing well-structured data forms. As genomic research advances, knowledge discovery from a large collection of scientific papers is becoming more important for efficient biological and biomedical researches. In this paper, we present a relation kernel based interaction extraction method to resolve this problem. We extract gene/protein interactions of Yeast (S.cerevisiae) from text documents with relation kernel. Kernel for relation extraction is constructed with predefined interaction corpus and set of interaction patterns. Proposed relation kernel for interaction extraction only exploits shallow parsed documents. Experimental results show that the proposed kernel method achieves a recall rate of 78.3% and precision rate of 79.9% for gene/protein interaction extraction without full parsing efforts.