Automated annotation of gene expression image sequences via non-parametric factor analysis and conditional random fields

AnnoFly: annotating Drosophila embryonic images based on an attention-enhanced RNN model

Motivation

In the post-genomic era, image-based transcriptomics have received huge attention, because the visualization of gene expression distribution is able to reveal spatial and temporal expression pattern, which is significantly important for understanding biological mechanisms. The Berkeley Drosophila Genome Project has collected a large-scale spatial gene expression database for studying Drosophila embryogenesis. Given the expression images, how to annotate them for the study of Drosophila embryonic development is the next urgent task. In order to speed up the labor-intensive labeling work, automatic tools are highly desired. However, conventional image annotation tools are not applicable here, because the labeling is at the gene-level rather than the image-level, where each gene is represented by a bag of multiple related images, showing a multi-instance phenomenon, and the image quality varies by image orientations and experiment batches. Moreover, different local regions of an image correspond to different CV annotation terms, i.e. an image has multiple labels. Designing an accurate annotation tool in such a multi-instance multi-label scenario is a very challenging task.

Results

To address these challenges, we develop a new annotator for the fruit fly embryonic images, called AnnoFly. Driven by an attention-enhanced RNN model, it can weight images of different qualities, so as to focus on the most informative image patterns. We assess the new model on three standard datasets. The experimental results reveal that the attention-based model provides a transparent approach for identifying the important images for labeling, and it substantially enhances the accuracy compared with the existing annotation methods, including both single-instance and multi-instance learning methods.

Availability and implementation

http://www.csbio.sjtu.edu.cn/bioinf/annofly/

Supplementary information

Supplementary data are available at Bioinformatics online.

New methods for computational decomposition of whole-mount in situ images enable effective curation of a large, highly redundant collection of Xenopus images

The precise anatomical location of gene expression is an essential component of the study of gene function. For most model organisms this task is usually undertaken via visual inspection of gene expression images by interested researchers. Computational analysis of gene expression has been developed in several model organisms, notably in Drosophila which exhibits a uniform shape and outline in the early stages of development. Here we address the challenge of computational analysis of gene expression in Xenopus, where the range of developmental stages of interest encompasses a wide range of embryo size and shape. Embryos may have different orientation across images, and, in addition, embryos have a pigmented epidermis that can mask or confuse underlying gene expression. Here we report the development of a set of computational tools capable of processing large image sets with variable characteristics. These tools efficiently separate the Xenopus embryo from the background, separately identify both histochemically stained and naturally pigmented regions within the embryo, and can sort images from the same gene and developmental stage according to similarity of gene expression patterns without information about relative orientation. We tested these methods on a large, but highly redundant, collection of 33,289 in situ hybridization images, allowing us to select representative images of expression patterns at different embryo orientations. This has allowed us to put a much smaller subset of these images into the public domain in an effective manner. The ‘isimage’ module and the scripts developed are implemented in Python and freely available on https://pypi.python.org/pypi/isimage/.

An important component of research into the function of genes in the developing organism is an understanding of both when and where the gene is expressed. Well established molecular techniques can be used to colour the embryo in regions where the gene of interest appears, and researchers will photograph such treated embryos at different stages of development to build up the story of the gene’s use. Small numbers of these expression pattern images may easily be examined by eye, but getting usable information from large collections of such images would take an enormous investment in time by trained scientists. Computational analysis is much to be preferred, but the task is complex and difficult to generalise. The frog Xenopus is an important model for studying vertebrate development, but up till now has had no purely computational methods available for analysing gene expression. Here we present a suite of computational tools based on a range of mathematical methods, capable of recognising the outline of the embryo against a variety of backgrounds, and within the embryo separately recognising areas of both gene expression and natural pigmentation. These tools work over a wide range of embryo shapes and imaging conditions, and, in our opinion, represent a major step towards full automation of anatomical gene expression annotation in vertebrate embryology.

Stability-driven nonnegative matrix factorization to interpret spatial gene expression and build local gene networks

Spatial gene expression patterns enable the detection of local covariability and are extremely useful for identifying local gene interactions during normal development. The abundance of spatial expression data in recent years has led to the modeling and analysis of regulatory networks. The inherent complexity of such data makes it a challenge to extract biological information. We developed staNMF, a method that combines a scalable implementation of nonnegative matrix factorization (NMF) with a new stability-driven model selection criterion. When applied to a set ofDrosophilaearly embryonic spatial gene expression images, one of the largest datasets of its kind, staNMF identified 21 principal patterns (PP). Providing a compact yet biologically interpretable representation ofDrosophilaexpression patterns, PP are comparable to a fate map generated experimentally by laser ablation and show exceptional promise as a data-driven alternative to manual annotations. Our analysis mapped genes to cell-fate programs and assigned putative biological roles to uncharacterized genes. Finally, we used the PP to generate local transcription factor regulatory networks. Spatially local correlation networks were constructed for six PP that span along the embryonic anterior-posterior axis. Using a two-tail 5% cutoff on correlation, we reproduced 10 of the 11 links in the well-studied gap gene network. The performance of PP with theDrosophiladata suggests that staNMF provides informative decompositions and constitutes a useful computational lens through which to extract biological insight from complex and often noisy gene expression data.

Bioimage Informatics in the context of Drosophila research

Modern biological research relies heavily on microscopic imaging. The advanced genetic toolkit of Drosophila makes it possible to label molecular and cellular components with unprecedented level of specificity necessitating the application of the most sophisticated imaging technologies. Imaging in Drosophila spans all scales from single molecules to the entire populations of adult organisms, from electron microscopy to live imaging of developmental processes. As the imaging approaches become more complex and ambitious, there is an increasing need for quantitative, computer-mediated image processing and analysis to make sense of the imagery. Bioimage Informatics is an emerging research field that covers all aspects of biological image analysis from data handling, through processing, to quantitative measurements, analysis and data presentation. Some of the most advanced, large scale projects, combining cutting edge imaging with complex bioimage informatics pipelines, are realized in the Drosophila research community. In this review, we discuss the current research in biological image analysis specifically relevant to the type of systems level image datasets that are uniquely available for the Drosophila model system. We focus on how state-of-the-art computer vision algorithms are impacting the ability of Drosophila researchers to analyze biological systems in space and time. We pay particular attention to how these algorithmic advances from computer science are made usable to practicing biologists through open source platforms and how biologists can themselves participate in their further development.