Multiclass cancer classification by support vector machines with class-wise optimized genes and probability estimates

We investigate the multiclass classification of cancer microarray samples. In contrast to classification of two cancer types from gene expression data, multiclass classification of more than two cancer types are relatively hard and less studied problem. We used class-wise optimized genes with corresponding one-versus-all support vector machine (OVA-SVM) classifier to maximize the utilization of selected genes. Final prediction was made by using probability scores from all classifiers. We used three different methods of estimating probability from decision value. Among the three probability methods, Platt's approach was more consistent, whereas, isotonic approach performed better for datasets with unequal proportion of samples in different classes. Probability based decision does not only gives true and fair comparison between different one-versus-all (OVA) classifiers but also gives the possibility of using them for any post analysis. Several ensemble experiments, an example of post analysis, of the three probability methods were implemented to study their effect in improving the classification accuracy. We observe that ensemble did help in improving the predictive accuracy of cancer data sets especially involving unbalanced samples. Four-fold external stratified cross-validation experiment was performed on the six multiclass cancer datasets to obtain unbiased estimates of prediction accuracies. Analysis of class-wise frequently selected genes on two cancer datasets demonstrated that the approach was able to select important and relevant genes consistent to literature. This study demonstrates successful implementation of the framework of class-wise feature selection and multiclass classification for prediction of cancer subtypes on six datasets.     

Bis-class: a new classification tool of methylation status using bayes classifier and local methylation information

Background

Whole genome sequencing of bisulfite converted DNA (‘methylC-seq’) method provides comprehensive information of DNA methylation. An important application of these whole genome methylation maps is classifying each position as a methylated versus non-methylated nucleotide. A widely used current method for this purpose, the so-called binomial method, is intuitive and straightforward, but lacks power when the sequence coverage and the genome-wide methylation level are low. These problems present a particular challenge when analyzing sparsely methylated genomes, such as those of many invertebrates and plants.

Results

We demonstrate that the number of sequence reads per position from methylC-seq data displays a large variance and can be modeled as a shifted negative binomial distribution. We also show that DNA methylation levels of adjacent CpG sites are correlated, and this similarity in local DNA methylation levels extends several kilobases. Taking these observations into account, we propose a new method based on Bayesian classification to infer DNA methylation status while considering the neighborhood DNA methylation levels of a specific site. We show that our approach has higher sensitivity and better classification performance than the binomial method via multiple analyses, including computational simulations, Area Under Curve (AUC) analyses, and improved consistencies across biological replicates. This method is especially advantageous in the analyses of sparsely methylated genomes with low coverage.

Conclusions

Our method improves the existing binomial method for binary methylation calls by utilizing a posterior odds framework and incorporating local methylation information. This method should be widely applicable to the analyses of methylC-seq data from diverse sparsely methylated genomes. Bis-Class and example data are provided at a dedicated website (http://bibs.snu.ac.kr/software/Bisclass).

Electronic supplementary material

The online version of this article (doi:10.1186/1471-2164-15-608) contains supplementary material, which is available to authorized users.     

Predicting the Risk of Suicide by Analyzing the Text of Clinical Notes

We developed linguistics-driven prediction models to estimate the risk of suicide. These models were generated from unstructured clinical notes taken from a national sample of U.S. Veterans Administration (VA) medical records. We created three matched cohorts: veterans who committed suicide, veterans who used mental health services and did not commit suicide, and veterans who did not use mental health services and did not commit suicide during the observation period (n = 70 in each group). From the clinical notes, we generated datasets of single keywords and multi-word phrases, and constructed prediction models using a machine-learning algorithm based on a genetic programming framework. The resulting inference accuracy was consistently 65% or more. Our data therefore suggests that computerized text analytics can be applied to unstructured medical records to estimate the risk of suicide. The resulting system could allow clinicians to potentially screen seemingly healthy patients at the primary care level, and to continuously evaluate the suicide risk among psychiatric patients.     

Potential distribution of Ursus americanus in Mexico and its persistence: Implications for conservation

The black bear Ursus americanus is an endangered species in Mexico. Its historical distribution has decreased by approximately 80% although its current distribution is not known with precision; it is only reported to be present in the mountains of Northern Mexico. This study proposes two ensemble models: Mexicós black bear (a) potential distribution compared with Natural Protected Areas (NPAs); and, (b) persistence areas for 2024. The current distribution variables are coniferous forest, elevation and dry forest. Suitable habitat for black bear (354,047 km², 18.07% of the country) was found mainly in the north of the Sonoran biogeographical zone, along the Sierra Madre Occidental, the center and south of the Sierra Madre Oriental and some northern regions of the Altiplano Norte. Comparing these areas with NPAs documented that only 12.41% of potential distribution coincided with current suitable habitat. There are unprotected areas in Sierra Madre Occidental center and central and southern of Sierra Madre Oriental. The model for 2024 indicates a reduction of suitable habitat of 64.5%, mainly in the northern Sonoran zone and the center Sierra Madre Occidental. On the other hand, areas that will persist (125,673 km²) are located along the two main mountain ranges of Mexico. Identification of these sites will allow strengthening of long-term conservation strategies.     

Whence tit-for-tat?

Summary In theoretical and empirical studies of the evolution of cooperation, the tit-for-tat strategy (i.e. cooperate unless your partner did not cooperate in the previous interaction) is widely considered to be of central importance. Nevertheless, surprisingly little is known about the conditions in which tit-for-tat appears and disappears across generations in a population of interacting individuals. Here, we apply a newly developed classifier-system model (EvA) in addressing this issue when the key features of interactions are caricatured using the iterated prisoner's dilemma game. Our simple representation of behavioural strategies as algorithms composed of two interacting rules allowed us to determine conditions in which tit-for-tat can replace noncooperative strategies and vice versa. Using direct game-theoretic analysis and simulations with the EvA model, we determined that no strategy is evolutionarily stable, but larger population sizes and longer sequences of interactions between individuals can yield transient dominance by tit-for-tat. Genetic drift among behaviourally equivalent strategies is the key mechanism underlying this dominance. Our analysis suggests that tit-for-tat could be important in nature for cognitively simple organisms of limited memory capacity, in strongly kin-selected or group-selected populations, when interaction sequences between individuals are relatively short, in moderate-sized populations of widely interacting individuals and when defectors appear in the population with moderate frequency.     

A Transcriptome-wide Translational Program Defined by LIN28B Expression Level

1. Download high-res image (155KB)
2. Download full-size image

     

复杂疾病驱使的融合SDA-SVM集成基因挖掘方法

提出了一种新颖的复杂疾病驱使的融合SDA-SVM(Stepwise Discriminant Analysis-Support Vector Machine,SDA-SVM)技术的集成基因挖掘方法。该集成方法融合逐步判别分析和支持向量机的优点,能够有效地进行复杂疾病相关基因的深度挖掘,使得挖掘出的基因能够较好地识别疾病类型和亚型。通过将该方法应用于一套弥散性大B细胞淋巴瘤DNA表达谱数据,并与其它基因挖掘方法对比,结果表明该方法挖掘出的基因具有较高的疾病相关性和较强的疾病类型识别能力。     

Interpretable gene expression classifier with an accurate and compact fuzzy rule base for microarray data analysis

An accurate classifier with linguistic interpretability using a small number of relevant genes is beneficial to microarray data analysis and development of inexpensive diagnostic tests. Several frequently used techniques for designing classifiers of microarray data, such as support vector machine, neural networks, k-nearest neighbor, and logistic regression model, suffer from low interpretabilities. This paper proposes an interpretable gene expression classifier (named iGEC) with an accurate and compact fuzzy rule base for microarray data analysis. The design of iGEC has three objectives to be simultaneously optimized: maximal classification accuracy, minimal number of rules, and minimal number of used genes. An "intelligent" genetic algorithm IGA is used to efficiently solve the design problem with a large number of tuning parameters. The performance of iGEC is evaluated using eight commonly-used data sets. It is shown that iGEC has an accurate, concise, and interpretable rule base (1.1 rules per class) on average in terms of test classification accuracy (87.9%), rule number (3.9), and used gene number (5.0). Moreover, iGEC not only has better performance than the existing fuzzy rule-based classifier in terms of the above-mentioned objectives, but also is more accurate than some existing non-rule-based classifiers.     

Predicting protein subcellular location by fusing multiple classifiers

One of the fundamental goals in cell biology and proteomics is to identify the functions of proteins in the context of compartments that organize them in the cellular environment. Knowledge of subcellular locations of proteins can provide key hints for revealing their functions and understanding how they interact with each other in cellular networking. Unfortunately, it is both time-consuming and expensive to determine the localization of an uncharacterized protein in a living cell purely based on experiments. With the avalanche of newly found protein sequences emerging in the post genomic era, we are facing a critical challenge, that is, how to develop an automated method to fast and reliably identify their subcellular locations so as to be able to timely use them for basic research and drug discovery. In view of this, an ensemble classifier was developed by the approach of fusing many basic individual classifiers through a voting system. Each of these basic classifiers was trained in a different dimension of the amphiphilic pseudo amino acid composition (Chou [2005] Bioinformatics 21: 10-19). As a demonstration, predictions were performed with the fusion classifier for proteins among the following 14 localizations: (1) cell wall, (2) centriole, (3) chloroplast, (4) cytoplasm, (5) cytoskeleton, (6) endoplasmic reticulum, (7) extracellular, (8) Golgi apparatus, (9) lysosome, (10) mitochondria, (11) nucleus, (12) peroxisome, (13) plasma membrane, and (14) vacuole. The overall success rates thus obtained via the resubstitution test, jackknife test, and independent dataset test were all significantly higher than those by the existing classifiers. It is anticipated that the novel ensemble classifier may also become a very useful vehicle in classifying other attributes of proteins according to their sequences, such as membrane protein type, enzyme family/sub-family, G-protein coupled receptor (GPCR) type, and structural class, among many others. The fusion ensemble classifier will be available at www.pami.sjtu.edu.cn/people/hbshen.     

Surface enhanced laser desorption ionization spectrometry reveals biomarkers for drug treatment but not dose

Here we describe the use of SELDI-MS to detect dose-dependent peptide changes in plasma from mice treated with vehicle or rosiglitazone at one of two doses (10 and 30 mg/kg). SELDI features differentiating spectra from the three conditions were found and used to train classifiers. Samples treated with vehicle could be reliably distinguished from samples treated with either dose, but samples treated with the different doses could not be reliably distinguished from one another. We conclude that while SELDI-TOF mass spectra can be used to distinguish treated from untreated samples, the reproducibility and information content of SELDI-TOF are currently not sufficient as a pharmacodynamic readout to distinguish between mice treated with 10 or 30 mg/kg of rosiglitazone. This raises more general questions about whether SELDI's sensitivity is sufficient for detecting dose-dependent changes in plasma.