Find open-source science resources

Variational Bayesian Multinomial Probit Regression with Gaussian Process Priors. It estimates class membership posterior probability employing variational and sparse approximation to the full posterior. This software also incorporates feature weighting by means of Automatic Relevance Determination.

Image segmentation is the process of identifying the borders of individual objects (in this case cells) within an image. This allows for the features of cells such as marker expression and morphology to be extracted, stored and analysed. simpleSeg provides functionality for user friendly, watershed based segmentation on multiplexed cellular images in R based on the intensity of user specified protein marker channels. simpleSeg can also be used for the normalization of single cell data obtained from multiple images.

This package implements methods to calculate information accretion for a given version of the gene ontology and uses this data to calculate remaining uncertainty, misinformation, and semantic similarity for given sets of predicted annotations and true annotations from a protein function predictor.

This package calculates probabilistic pathway scores using gene expression data. Gene expression values are aggregated into pathway-based scores using Bayesian network representations of biological pathways.

Reads files exported from 'QX Manager or QuantaSoft' containing amplitude values from a run of ddPCR (96 well plate) and robustly sets thresholds to determine positive droplets for each channel of each individual well. Concentration and normalized concentration in addition to other metrics is then calculated for each well. Results are returned as a table, optionally written to file, as well as optional plots (scatterplot and histogram) for both channels per well written to file. The package includes a shiny application which provides an interactive and user-friendly interface to the full functionality of PoDCall.

PAA imports single color (protein) microarray data that has been saved in gpr file format - esp. ProtoArray data. After preprocessing (background correction, batch filtering, normalization) univariate feature preselection is performed (e.g., using the "minimum M statistic" approach - hereinafter referred to as "mMs"). Subsequently, a multivariate feature selection is conducted to discover biomarker candidates. Therefore, either a frequency-based backwards elimination aproach or ensemble feature selection can be used. PAA provides a complete toolbox of analysis tools including several different plots for results examination and evaluation.

OCTAD provides a platform for virtually screening compounds targeting precise cancer patient groups. The essential idea is to identify drugs that reverse the gene expression signature of disease by tamping down over-expressed genes and stimulating weakly expressed ones. The package offers deep-learning based reference tissue selection, disease gene expression signature creation, pathway enrichment analysis, drug reversal potency scoring, cancer cell line selection, drug enrichment analysis and in silico hit validation. It currently covers ~20,000 patient tissue samples covering 50 cancer types, and expression profiles for ~12,000 distinct compounds.

This package provides uniform interfaces to machine learning code for data in R and Bioconductor containers.

The MicrobiomeExplorer R package is designed to facilitate the analysis and visualization of marker-gene survey feature data. It allows a user to perform and visualize typical microbiome analytical workflows either through the command line or an interactive Shiny application included with the package. In addition to applying common analytical workflows the application enables automated analysis report generation.

Provide functions for performing abundance and compositional based binning on metagenomic samples, directly from FASTA or FASTQ files. Functions are implemented in Java and called via rJava. Parallel implementation that operates directly on input FASTA/FASTQ files for fast execution. Inputs may be file paths or Biostrings/ShortRead sequence objects; results are returned as a MetabinResult S4 object wrapping cluster assignments, algorithm parameters, and input metadata.

Here we present Link-HD, an approach to integrate heterogeneous datasets, as a generalization of STATIS-ACT (“Structuration des Tableaux A Trois Indices de la Statistique–Analyse Conjointe de Tableaux”), a family of methods to join and compare information from multiple subspaces. However, STATIS-ACT has some drawbacks since it only allows continuous data and it is unable to establish relationships between samples and features. In order to tackle these constraints, we incorporate multiple distance options and a linear regression based Biplot model in order to stablish relationships between observations and variable and perform variable selection.

This package provides functions for calculation and visualization of performance metrics for evaluation of ranking and binary classification (assignment) methods. Various types of performance plots can be generated programmatically. The package also contains a shiny application for interactive exploration of results.

Classification using generalized partial least squares for two-group and multi-group (more than 2 group) classification.

Statistic methods to evaluate variations of differential expression (DE) between multiple biological conditions. It takes into account the fold-changes and p-values from previous differential expression (DE) results that use large-scale data (*e.g.*, microarray and RNA-seq) and evaluates which genes would react in response to the distinct experiments. This evaluation involves an unique pipeline of statistical methods, including weighted summarization, quantile detection, cluster analysis, and ANOVA tests, in order to classify a subset of relevant genes whose DE is similar or dependent to certain biological factors.

Comprehensive package to automatically train and validate a multi-class SVM classifier based on gene expression data. Provides transparent selection of gene markers, their coexpression networks, and an interface to query the classifier.

Feature selection aims to identify and remove redundant, irrelevant and noisy variables from high-dimensional datasets. Selecting informative features affects the subsequent classification and regression analyses by improving their overall performances. Several methods have been proposed to perform feature selection: most of them relies on univariate statistics, correlation, entropy measurements or the usage of backward/forward regressions. Herein, we propose an efficient, robust and fast method that adopts stochastic optimization approaches for high-dimensional. GARS is an innovative implementation of a genetic algorithm that selects robust features in high-dimensional and challenging datasets.

fastseg implements a very fast and efficient segmentation algorithm. It has similar functionality as DNACopy (Olshen and Venkatraman 2004), but is considerably faster and more flexible. fastseg can segment data from DNA microarrays and data from next generation sequencing for example to detect copy number segments. Further it can segment data from RNA microarrays like tiling arrays to identify transcripts. Most generally, it can segment data given as a matrix or as a vector. Various data formats can be used as input to fastseg like expression set objects for microarrays or GRanges for sequencing data. The segmentation criterion of fastseg is based on a statistical test in a Bayesian framework, namely the cyber t-test (Baldi 2001). The speed-up arises from the facts, that sampling is not necessary in for fastseg and that a dynamic programming approach is used for calculation of the segments' first and higher order moments.

This package assists in demultiplexing scRNAseq data using both cell hashing and SNPs data. The SNP profile of each group os learned using high confidence assignments from the cell hashing data. Cells which cannot be assigned with high confidence from the cell hashing data are assigned to their most similar group based on their SNPs. We also provide some helper function to optimise SNP selection, create training data and merge SNP data into the SingleCellExperiment framework.

This package implements four major subtype classifiers for high-grade serous (HGS) ovarian cancer as described by Helland et al. (PLoS One, 2011), Bentink et al. (PLoS One, 2012), Verhaak et al. (J Clin Invest, 2013), and Konecny et al. (J Natl Cancer Inst, 2014). In addition, the package implements a consensus classifier, which consolidates and improves on the robustness of the proposed subtype classifiers, thereby providing reliable stratification of patients with HGS ovarian tumors of clearly defined subtype.

A method that allows for the use of a collection of non-matched normal tissue samples. Our approach uses a non-parametric bootstrap subsampling of the available reference samples to estimate the distribution of read counts from targeted sequencing. As inspired by random forest, this is combined with a procedure that subsamples the amplicons associated with each of the targeted genes. The obtained information allows us to reliably classify the copy number aberrations on the gene level.

This package provides a comprehensive collection of various microarray-based classification algorithms both from Machine Learning and Statistics. Variable Selection, Hyperparameter tuning, Evaluation and Comparison can be performed combined or stepwise in a user-friendly environment.

Package for modified nearest-neighbor classification based on calculation of a similarity threshold distinguishing within-group from between-group comparisons.

The software formalises a framework for classification and survival model evaluation in R. There are four stages; Data transformation, feature selection, model training, and prediction. The requirements of variable types and variable order are fixed, but specialised variables for functions can also be provided. The framework is wrapped in a driver loop that reproducibly carries out a number of cross-validation schemes. Functions for differential mean, differential variability, and differential distribution are included. Additional functions may be developed by the user, by creating an interface to the framework.

CHETAH (CHaracterization of cEll Types Aided by Hierarchical classification) is an accurate, selective and fast scRNA-seq classifier. Classification is guided by a reference dataset, preferentially also a scRNA-seq dataset. By hierarchical clustering of the reference data, CHETAH creates a classification tree that enables a step-wise, top-to-bottom classification. Using a novel stopping rule, CHETAH classifies the input cells to the cell types of the references and to "intermediate types": more general classifications that ended in an intermediate node of the tree.

Defining the identity of a cell is fundamental to understand the heterogeneity of cells to various environmental signals and perturbations. We present Cepo, a new method to explore cell identities from single-cell RNA-sequencing data using differential stability as a new metric to define cell identity genes. Cepo computes cell-type specific gene statistics pertaining to differential stable gene expression.

Feature selection is critical in omics data analysis to extract restricted and meaningful molecular signatures from complex and high-dimension data, and to build robust classifiers. This package implements a new method to assess the relevance of the variables for the prediction performances of the classifier. The approach can be run in parallel with the PLS-DA, Random Forest, and SVM binary classifiers. The signatures and the corresponding 'restricted' models are returned, enabling future predictions on new datasets. A Galaxy implementation of the package is available within the Workflow4metabolomics.org online infrastructure for computational metabolomics.