Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/070904

OrchID: a Generalized Framework for Taxonomic Classification of Images Using Evolved Artificial Neural Networks

Serano Pereira·Barbara Gravendeel·Patrick Wijntjes·Rutger Vos
View original source ↗Export metadata ↓

Abstract

Taxonomic expertise for the identification of species is rare and costly. On-going advances in computer vision and machine learning have led to the development of numerous semi- and fully automated species identification systems. However, these systems are rarely agnostic to specific morphology, rarely can perform taxonomic \"approximation\" (by which we mean partial identification at least to higher taxonomic level if not to species), and frequently rely on costly scientific imaging technologies. We present a generic, hierarchical identification system for automated taxonomic approximation of organisms from images. We assessed the effectiveness of this system using photographs of slipper orchids (Cypripedioideae), for which we implemented image pre-processing, segmentation, and colour and shape feature extraction algorithms to obtain digital phenotypes for 116 species. The identification system trained on these digital phenotypes uses a nested hierarchy of artificial neural networks for pattern recognition and automated classification that mirrors the Linnean taxonomy, such that user-submitted photos can be assigned a genus, section, and species classification by traversing this hierarchy. Performance of the identification system varied depending on photo quality, number of species included for training, and desired taxonomic level for identification. High quality photos were scarce for some taxa and were under-represented in the training set, resulting in imbalanced network training. The image features used for training were sufficient to reliably identify photos to the correct genus but less so to the correct section and species. The outcomes of this project include a library of feature extraction algorithms called ImgPheno, a collection of scripts for neural network training called NBClassify, a library for evolutionary optimization of artificial neural network construction called AI::FANN::Evolving and a planned web application called OrchID for identification of user-submitted images. All project outcomes are open source and freely available.

Explore related subjects

BioinformaticsPreprint

Keep this discovery

BibTeXRIS

Serano Pereira, Barbara Gravendeel, Patrick Wijntjes, Rutger Vos. 2016-08-22. OrchID: a Generalized Framework for Taxonomic Classification of Images Using Evolved Artificial Neural Networks. https://doi.org/10.1101/070904

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2016

iCARE: An R Package to Build and Apply Absolute Risk Models

This report describes a R package, called the Individualized Coherent Absolute Risk Estimation (iCARE) tool, that allows researchers to build and evaluate models for absolute risk and apply them to estimate an individuals risk of developing disease during a specified time interval based on a set of user defined input parameters. An attractive feature of the software is that it gives users flexibility to update models rapidly based on new knowledge on risk factors and tailor models to different populations by specifying three input arguments: (1) a model for relative risk, (2) an age-specific disease incidence rate, (3) the distribution of risk factors for the population of interest. The tool can handle missing information on risk factors for individuals for whom risks are to be predicted using a coherent approach where all estimates are derived from a single model after appropriate model averaging. The software allows single nucleotide polymorphisms (SNPs) to be incorporated into the model using published odds ratios and allele frequencies. The validation component of the software implements the methods for evaluation of model calibration, discrimination and risk-stratification based on independent validation datasets. We provide an illustration of the utility of iCARE for building, validating and applying absolute risk models using breast cancer as an example.

Bioinformatics↗
bioRxiv2016

deSPI: efficient classification of metagenomic reads with lightweight de Bruijn graph-based reference indexing

SummaryIn metagenomic studies, fast and effective tools are on wide demand to implement taxonomy classification for upto billions of reads. Herein, we propose deSPI, a novel read classification method that classifies reads by recognizing and analyzing the matches between reads and reference with de Bruijn graph-based lightweight reference indexing. deSPI has faster speed with relatively small memory footprint, meanwhile, it can also achieve higher or similar sensitivity and accuracy.\n\nAvailabilitythe C++ source code of deSPI is available at https://github.com/hitbc/deSPI\n\nContactydwang@hit.edu.cn\n\nSupplementary informationSupplementary data are available at Bioinformatics online.

Bioinformatics↗
bioRxiv2016

De novo assembly of viral quasispecies using overlap graphs

A viral quasispecies, the ensemble of viral strains populating an infected person, can be highly diverse. For optimal assessment of virulence, pathogenesis and therapy selection, determining the haplotypes of the individual strains can play a key role. As many viruses are subject to high mutation and recombination rates, high-quality reference genomes are often not available at the time of a new disease outbreak. We present SAVAGE, a computational tool for reconstructing individual haplotypes of intrahost virus strains without the need for a high-quality reference genome. SAVAGE makes use of either FM-index based data structures or ad-hoc consensus reference sequence for constructing overlap graphs from patient sample data. In this overlap graph, nodes represent reads and/or contigs, while edges reflect that two reads/contigs, based on sound statistical considerations, represent identical haplotypic sequence. Following an iterative scheme, a new overlap assembly algorithm that is based on the enumeration of statistically well-calibrated groups of reads/contigs then efficiently reconstructs the individual haplotypes from this overlap graph. In benchmark experiments on simulated and on real deep coverage data, SAV-AGE drastically outperforms generic de novo assemblers as well as the only specialized de novo viral quasispecies assembler available so far. When run on ad-hoc consensus reference sequence, SAVAGE performs very favorably in comparison with state-of-the-art reference genome guided tools. We also apply SAVAGE on two deep coverage samples of patients infected by the Zika and the hepatitis C virus, respectively, which sheds light on the genetic structures of the respective viral quasispecies.

Bioinformatics↗