Ekrem Varoglu

Vote-Based Classifier Selection for Biomedical NER Using Genetic Algorithms

We propose a genetic algorithm for constructing a classifier ensemble using a vote-based classifier selection approach for biomedical named entity recognition task. Assuming that the reliability of the predictions of each classifier differs among classes, the proposed approach is based on dynamic selection of the classifiers by taking into account their individual votes. During testing, the classifiers whose votes are considered as being reliable are combined using weighted majority voting. The classifier ensemble formed by the proposed scheme surpasses the full object F-score of the best individual classifier and the ensemble of all classifiers by 2.5% and 1.3% respectively.

Recognizing biomedical named entities using SVMs: improving recognition performance with a minimal set of features

In this paper, Support Vector Machines (SVMs) are applied to the identification and automatic annotation of biomedical named entities in the domain of molecular biology, as an extension of the traditional named entity recognition task to special domains. The effect of the use of well-known features such as word formation patterns, lexical, morphological, and surface words on recognition performance is investigated. Experiments have been conducted using the train and test data made public at the Bio-Entity Recognition Task at JNLPBA 2004. An F-score of 69.87% was obtained by using a carefully selected combination of a minimal set of features, which can be easily computed from training data without any use of post-processing or external resources.

ChemTok: A New Rule Based Tokenizer for Chemical Named Entity Recognition

Named Entity Recognition (NER) from text constitutes the first step in many text mining applications. The most important preliminary step for NER systems using machine learning approaches is tokenization where raw text is segmented into tokens. This study proposes an enhanced rule based tokenizer, ChemTok, which utilizes rules extracted mainly from the train data set. The main novelty of ChemTok is the use of the extracted rules in order to merge the tokens split in the previous steps, thus producing longer and more discriminative tokens. ChemTok is compared to the tokenization methods utilized by ChemSpot and tmChem. Support Vector Machines and Conditional Random Fields are employed as the learning algorithms. The experimental results show that the classifiers trained on the output of ChemTok outperforms all classifiers trained on the output of the other two tokenizers in terms of classification performance, and the number of incorrectly segmented entities.

Face modeling and adaptive texture mapping for model based video coding

3D facial synthesis has been frequently used in model based video coding applications and became popular in various multimedia applications. In this paper a 3D face model, its adaptation algorithm and a texture mapping method using two orthogonal photos are presented to solve several 3D estimation problems in model based video coding. We are successfully estimating the frames between the front and the side views of the face. The experimental results show that the proposed Rotation Adaptive Texture Mapping (RATM) technique increases the visual quality of the synthesized face during rotations of the head, while achieving a PSNR value up to 33dB.

A symmetric term weighting scheme for text categorization based on term occurrence probabilities

Term weighting schemes used in text categorization can be considered as functions of term occurence probabilities in positive and negative classes. In this paper, widely used weighting schemes are firstly evaluated from this perspective. Then, a novel feature weighting scheme based on term occurrence probabilities is proposed. Experiments conducted using SVM classifier on the Reuters-21578 ModApte Top10 dataset shows that the proposed method outperforms other well known measures such as CHI, IG, OR and RF in terms of macro-F1 and micro-F1 scores.

Functional variation of alternative splice forms in their protein interaction networks: a literature mining approach

Analyzing protein interactions and protein functions iscrucial for the analysis of complex biological processesas well as the consequences from aberrant gene pro-ducts [1]. Protein-Protein Interaction Networks (PPIN)are invaluable means enabling scientists to get a globalunderstanding of interactomes, while analyzing indivi-dual protein functions [2]. High-throughput experimentsand complete literature mining analyses have been usedto deliver well-structured data into scientific databasesreporting on proteins, their interactions and functions.These repositories form a precious resource to scientists,butonlycoveraportionoftheproteomeandoftenunderrepresented alternative splice forms [3].Alternative splicing (AS) is a cellular process that pro-duces from a single gene different physical variants of agiven protein which may differ in its structure or itsfunction. This process produces molecular variabilityand contributes to the complexity of the proteomes andtheir interactomes. The analysis of AS shows that thisprocess is most relevant to molecular regulation pro-cesses. In this research, we attempt to identify functionalvariability linked to alternative splice forms within theirPPINs from the scientific literature. For this purpose, wegather AS events and analyze the transcript data for16,826 different genes from the HumanSDB3 database[4,5]. We have collected around 4 million abstracts fromNCBI’sPubMedbyutilizingarichsearchtermsetforeach individual isoform by using Gene DB, SwissprotDB and synonym generation. We then utilize an SVMclassifier which uses in-domain features together withstandard term weights and have trained it on theBioCreative-II IAS corpus (81.31% F

Balanced undersampling: a novel sentence-based undersampling method to improve recognition of named entities in chemical and biomedical text

The class imbalance problem is a key factor that affects the performance of many classification tasks when using machine learning methods. This mainly refers to the problem where the number of samples in certain classes is much greater than in others. Such imbalance considerably affects the performance of classifiers in which the majority class or classes are often favored, thus resulting in high-precision/low-recall classifiers. Named entity recognition in free text suffers from this problem to a large extent because in any given free text, many samples do not belong to a specific entity. Furthermore, the data used in this specific type of classification is in sequenced mode and is different than that used in other common classification tasks such as image classification, spam detection, and text classification in which no semantic or syntactic relation exists between samples. In this study, we propose an undersampling approach for sequenced data that preserves existing correlations between sequenced samples that comprise sentences and thus improve the performance of classifiers. We call this method balanced undersampling (BUS). Considering the recent increased interest in the use of NER in the chemical and biomedical domains, the proposed method is developed and tested on four recent state-of-the-art corpora in these domains, including BioCreative IV ChemDNER, Bio-entity Recognition Challenge of JNLPBA (JNLPBA), SemEval2013 DDI DrugBank, and SemEval2013 DDI Medline datasets. The performance of the proposed method is evaluated against two other common undersampling methods: random undersampling and stop-word filtering. Our method is shown to outperform both methods with respect to F-score for all datasets used.

Role of neurotransmitter receptors in behavioral disorders - a high-throughput analysis using text mining

Allelic variation in neurotransmitter receptors have been shown to be implicated both in behavioural variations across individuals in a given population and in various behavioural disorders. There are two aspects of synaptic neurotransmission and its implications in behavioural disorders, both of which are important in healthcare management for such conditions. Firstly, certain allelic variations lead to an increased susceptibility to certain behavioural disorders. Secondly, specific allelic variations determine the response of affected individuals to available drug treatment options. These studies are in general done on molecular level and focused on single or a few disorders at a time. In this study, we aim to approach the relationship of neurotransmitter receptors to behavioural disorders from a different perspective. We employ state-of-the-art text mining methods to put together a comprehensive database linking receptors with specific disorders. This tool will enable researchers in the field to have easy access to a large amount of neurotransmission and disease data, and analyze these conditions within a larger scope.

Neurotransmitter receptor genotypes associated with mental and behavioral disorders

AIM Investigation of association studies within the field of mental and behavioral disorders is of value given their complex molecular etiology including epistatic interactions of multiple genes with small effects. MATERIALS & METHODS Utilizing biomedical text mining, associations are uncovered for all mental and behavioral conditions listed in Diagnostic and Statistical Manual of Mental Disorders Text Revision. Specifically, a computational pipeline is designed to retrieve neurotransmitter receptor variations from biomedical literature with a text mining approach, where unique polymorphisms are also mined. RESULTS Analyses of 1337 unique neurotransmitter receptors and 465 distinct conditions yield 1568 unique gene-disease associations. CONCLUSION This study takes an unconventional approach to association studies and generates a novel dataset of associations for disorders such as major depression and schizophrenia, which provides a global perspective for their genetic etiology.