Search

Abstract The Structural Genomics Consortium is an international open science research organization with a focus on accelerating early-stage drug discovery, namely hit discovery and optimization. We, as many others, believe that artificial intelligence (AI) is poised to be a main accelerator in the field. The question is then how to best benefit from recent advances in AI and how to generate, format and disseminate data to enable future breakthroughs in AI-guided drug discovery. We present here the recommendations of a working group composed of experts from both the public and private sectors. Robust data management requires precise ontologies and standardized vocabulary while a centralized database architecture across laboratories facilitates data integration into high-value datasets. Lab automation and opening electronic lab notebooks to data mining push the boundaries of data sharing and data modeling. Important considerations for building robust machine-learning models include transparent and reproducible data processing, choosing the most relevant data representation, defining the right training and test sets, and estimating prediction uncertainty. Beyond data-sharing, cloud-based computing can be harnessed to build and disseminate machine-learning models. Important vectors of acceleration for hit and chemical probe discovery will be (1) the real-time integration of experimental data generation and modeling workflows within design-make-test-analyze (DMTA) cycles openly, and at scale and (2) the adoption of a mindset where data scientists and experimentalists work as a unified team, and where data science is incorporated into the experimental design.

Organised data is easy to use but the rapid developments in the field of bioimaging, with improvements in instrumentation, detectors, software and experimental techniques, have resulted in an explosion of the volumes of data being generated, making well-organised data an elusive goal. This guide offers a handful of recommendations for bioimage depositors, analysts and microscope and software developers, whose implementation would contribute towards better organised data in preparation for archival. Based on our experience archiving large image datasets in EMPIAR, the BioImage Archive and BioStudies, we propose a number of strategies that we believe would improve the usability (clarity, orderliness, learnability, navigability, self-documentation, coherence and consistency of identifiers, accessibility, succinctness) of future data depositions more useful to the bioimaging community (data authors and analysts, researchers, clinicians, funders, collaborators, industry partners, hardware/software producers, journals, archive developers as well as interested but non-specialist users of bioimaging data). The recommendations that may also find use in other data-intensive disciplines. To facilitate the process of analysing data organisation, we present bandbox, a Python package that provides users with an assessment of their data by flagging potential issues, such as redundant directories or invalid characters in file or folder names, that should be addressed before archival. We offer these recommendations as a starting point and hope to engender more substantial conversations across and between the various data-rich communities.

Paraneoplastic syndromes (PS) are uncommon and are known to mimic other clinical entities, often carrying significant morbidity and mortality. The commonest cause of extra-ocular muscle enlargement (EOME) is thyroid eye disease (TED). Rarely, PS can cause EOME and masquerade as TED. We describe a 52-year-old female who presented with diarrhoea, acute kidney injury and electrolyte imbalance. An ophthalmic review identified right upper lid retraction. MRI orbits showed increased thickness of the inferior and medial recti bilaterally, presumed as TED. Whilst investigating her diarrhoea, imaging revealed a large rectosigmoid tumour which required surgical excision. In the context of electrolyte disturbance and acute kidney injury, a diagnosis of McKittrick–Wheelock syndrome (MWS) was made. Following successful surgery, electrolyte imbalance, diarrhoea and eyelid retraction improved. Repeat MRI orbits displayed complete resolution of EOME. To our knowledge, this is the first case of MWS presenting with PS-EOME masquerading as TED.

Quantitative gene regulation at the cell population level can be achieved by two fundamentally different modes of regulation at individual gene copies. A ‘digital’ mode involves binary ON/OFF expression states, with population-level variation arising from the proportion of gene copies in each state, while an ‘analog’ mode involves graded expression levels at each gene copy. At the Arabidopsis floral repressor FLOWERING LOCUS C (FLC), ‘digital’ Polycomb silencing is known to facilitate quantitative epigenetic memory in response to cold. However, whether FLC regulation before cold involves analog or digital modes is unknown. Using quantitative fluorescent imaging of FLC mRNA and protein, together with mathematical modeling, we find that FLC expression before cold is regulated by both analog and digital modes. We observe a temporal separation between the two modes, with analog preceding digital. The analog mode can maintain intermediate expression levels at individual FLC gene copies, before subsequent digital silencing, consistent with the copies switching OFF stochastically and heritably without cold. This switch leads to a slow reduction in FLC expression at the cell population level. These data present a new paradigm for gradual repression, elucidating how analog transcriptional and digital epigenetic memory pathways can be integrated.

Crossover numbers and positions are tightly controlled but the mechanism involved is still obscure. Here, the authors, using quantitative super-resolution cytogenetics and mathematical modelling, show that diffusion mediated coarsening of HEI10, an E3-ligase domain containing protein, may explain meiotic crossover positioning in Arabidopsis.

We present the MATLAB code Spirality, a novel method for measuring spiral arm pitch angles by fitting galaxy images to spiral templates of known pitch. Computation time is typically on the order of 2 min per galaxy, assuming 8 GB of working memory. We tested the code using 117 synthetic spiral images with known pitches, varying both the spiral properties and the input parameters. The code yielded correct results for all synthetic spirals with galaxy-like properties. We also compared the code’s results to two-dimensional Fast Fourier Transform (2DFFT) measurements for the sample of nearby galaxies defined by DMS PPak. Spirality’s error bars overlapped 2DFFT’s error bars for 26 of the 30 galaxies. The two methods’ agreement correlates strongly with galaxy radius in pixels and also with i-band magnitude, but not with redshift, a result that is consistent with at least some galaxies’ spiral structure being fully formed by z=1.2, beyond which there are few galaxies in our sample. The Spirality code package also includes GenSpiral, which produces FITS images of synthetic spirals, and SpiralArmCount, which uses a one-dimensional Fast Fourier Transform to count the spiral arms of a galaxy after its pitch is determined. All code is freely available.

The histone modification H3K27me3 plays a central role in Polycomb-mediated epigenetic silencing. H3K27me3 recruits and allosterically activates Polycomb Repressive Complex 2 (PRC2), which adds this modification to nearby histones, providing a read/write mechanism for inheritance through DNA replication. However, for some PRC2 targets, a purely histone-based system for epigenetic inheritance may be insufficient. We address this issue at the Polycomb target FLOWERING LOCUS C (FLC) in Arabidopsis thaliana, as a narrow nucleation region of only ~three nucleosomes within FLC mediates epigenetic state switching and subsequent memory over many cell cycles. To explain the memory’s unexpected persistence, we introduce a mathematical model incorporating extra protein memory storage elements with positive feedback that persist at the locus through DNA replication, in addition to histone modifications. Our hybrid model explains many features of epigenetic switching/memory at FLC and encapsulates generic mechanisms that may be widely applicable.

Summary: Deep learning, a set of approaches using artificial neural networks, has generated rapid recent advancements in machine learning. Deep learning does, however, have the potential to reduce the reproducibility of scientific results. Model outputs are critically dependent on the data and processing approach used to initially generate the model, but this provenance information is usually lost during model training. To avoid a future reproducibility crisis, we need to improve our deep-learning model management. The FAIR principles for data stewardship and software/workflow implementation give excellent high-level guidance on ensuring effective reuse of data and software. We suggest some specific guidelines for the generation and use of deep-learning models in science and explain how these relate to the FAIR principles. We then present dtoolAI, a Python package that we have developed to implement these guidelines. The package implements automatic capture of provenance information during model training and simplifies model distribution. The Bigger Picture: Science has made use of machine learning, a way of teaching computers to understand patterns in data, for a long time. Deep learning, based on the way that real brains process data, has brought enormous improvements in the speed and accuracy of image and language processing over the last few years. However, the “black box” nature of deep-learning models makes scientific analyses that make use of them difficult to reproduce.In this work, we show how we might be able to improve long-term reproducibility for data analyses that rely on deep-learning models. We do this by giving guidance on how specific aspects of the FAIR principles for data management can be applied to training and using these models. We also present dtoolAI, a software tool and code library we have developed. We hope that in the future, adoption of our guidelines or similar principles will improve our collective trust in results that arise from deep learning.

The explosion in volumes and types of data has led to substantial challenges in data management. These challenges are often faced by front-line researchers who are already dealing with rapidly changing technologies and have limited time to devote to data management. There are good high-level guidelines for managing and processing scientific data. However, there is a lack of simple, practical tools to implement these guidelines. This is particularly problematic in a highly distributed research environment where needs differ substantially from group to group and centralised solutions are difficult to implement and storage technologies change rapidly. To meet these challenges we have developed dtool, a command line tool for managing data. The tool packages data and metadata into a unified whole, which we call a dataset. The dataset provides consistency checking and the ability to access metadata for both the whole dataset and individual files. The tool can store these datasets on several different storage systems, including a traditional file system, object store (S3 and Azure) and iRODS. It includes an application programming interface that can be used to incorporate it into existing pipelines and workflows. The tool has provided substantial process, cost, and peace-of-mind benefits to our data management practices and we want to share these benefits. The tool is open source and available freely online at http://dtool.readthedocs.io.

Methods that allow the study of gene expression regulation are continually advancing. Here, we present an in situ hybridization protocol capable of detecting individual mRNA molecules in plant root cells, thus permitting the accurate quantification and localization of mRNA within fixed samples (Duncan et al., 2016; Rosa et al., 2016). This single molecule RNA fluorescence in situ hybridization (smFISH) uses multiple single-labelled oligonucleotide probes to bind target RNAs and generate diffraction-limited signals that can be detected using a wide-field fluorescence microscope. We adapted a recent version of this method that uses 48 fluorescently labeled DNA oligonucleotides (20 mers) to hybridize to different portions of each transcript (Raj et al., 2008). This approach is simple to implement and has the advantage that it can be readily applied to any genetic background.

There has been steady improvement in methods for capturing bioimages. However analysing these images still remains a challenge. The Python programming language provides a powerful and flexible environment for scientific computation. It has a wide range of supporting libraries for image processing but lacks native support for common bioimage formats, and requires specific code to be written to ensure that suitable audit trails are generated and analyses are reproducible. Here we describe the development of a Python tool that: (1) allows users to quickly view and explore microscopy data; (2) generate reproducible analyses, encoding a complete history of image transformations from raw data to final result; and (3) scale up analyses from initial exploration to high throughput processing pipelines, with a minimal amount of extra effort. The tool, jicbioimage, is open source and freely available online at http://jicbioimage.readthedocs.io.

Inheritance of gene expression states is fundamental for cells to ‘remember’ past events, such as environmental or developmental cues. The conserved Polycomb Repressive Complex 2 (PRC2) maintains epigenetic repression of many genes in animals and plants and modifies chromatin at its targets. Histones modified by PRC2 can be inherited through cell division. However, it remains unclear whether this inheritance can direct long-term memory of individual gene expression states (cis memory) or instead if local chromatin states are dictated by the concentrations of diffusible factors (trans memory). By monitoring the expression of two copies of the Arabidopsis Polycomb target gene FLOWERING LOCUS C (FLC) in the same plants, we show that one copy can be repressed while the other is active. Furthermore, this ‘mixed’ expression state is inherited through many cell divisions as plants develop. These data demonstrate that epigenetic memory of FLC expression is stored not in trans but in cis.

Organised data is easy to use but the rapid developments in the field of bioimaging, with improvements in instrumentation, detectors, software and experimental techniques, have resulted in an explosion of the volumes of data being generated, making well-organised data an elusive goal. This guide offers a handful of recommendations for bioimage depositors, analysts and microscope and software developers, whose implementation would contribute towards better organised data in preparation for archival. Based on our experience archiving large image datasets in EMPIAR, the BioImage Archive and BioStudies, we propose a number of strategies that we believe would improve the usability (clarity, orderliness, learnability, navigability, self-documentation, coherence and consistency of identifiers, accessibility, succinctness) of future data depositions more useful to the bioimaging community (data authors and analysts, researchers, clinicians, funders, collaborators, industry partners, hardware/software producers, journals, archive developers as well as interested but non-specialist users of bioimaging data). The recommendations that may also find use in other data-intensive disciplines. To facilitate the process of analysing data organisation, we present bandbox, a Python package that provides users with an assessment of their data by flagging potential issues, such as redundant directories or invalid characters in file or folder names, that should be addressed before archival. We offer these recommendations as a starting point and hope to engender more substantial conversations across and between the various data-rich communities.

Organised data is easy to use but the growth of bioimaging, with improvements in instrumentation, detectors, software and experimental techniques has resulted in an explosion in the volumes of data being generated, making this an elusive goal. This guide offers a handful of recommendations whose implementation would contribute towards better organised data in preparation for archival. Based on our experience archiving large image datasets in EMPIAR, the BioImage Archive and BioStudies, we propose a number of strategies that we believe would make future data depositions more useful to the bioimaging community and that may also find use in other data-intensive disciplines. To facilitate the process of analysing data organisation, we present bandbox, a Python package that provides users with an assessment of their data by flagging potential issues that could be addressed before archival.