A deep learning approach for staging embryonic tissue isolates with small data

Abstract

Machine learning approaches are becoming increasingly widespread and are now present in most areas of research. Their recent surge can be explained in part due to our ability to generate and store enormous amounts of data with which to train these models. The requirement for large training sets is also responsible for limiting further potential applications of machine learning, particularly in fields where data tend to be scarce such as developmental biology. However, recent research seems to indicate that machine learning and Big Data can sometimes be decoupled to train models with modest amounts of data. In this work we set out to train a CNN-based classifier to stage zebrafish tail buds at four different stages of development using small information-rich data sets. Our results show that two and three dimensional convolutional neural networks can be trained to stage developing zebrafish tail buds based on both morphological and gene expression confocal microscopy images, achieving in each case up to 100% test accuracy scores. Importantly, we show that high accuracy can be achieved with data set sizes of under 100 images, much smaller than the typical training set size for a convolutional neural net. Furthermore, our classifier shows that it is possible to stage isolated embryonic structures without the need to refer to classic developmental landmarks in the whole embryo, which will be particularly useful to stage 3D culture in vitro systems such as organoids. We hope that this work will provide a proof of principle that will help dispel the myth that large data set sizes are always required to train CNNs, and encourage researchers in fields where data are scarce to also apply ML approaches.

Accurate staging of chick embryonic tissues via deep learning

“…Previous efforts to classify microscopy images in developmental biology have focused on hyperparameter optimisation (7) without data augmentation. In contrast, we have instead focused on exploring different data augmentations, as we can fully exploit biological domain expertise when augmenting data, as opposed to hyperparameter optimisation.…”

“…The employment of DNNs in analyses of in vitro fertilisation (8) and cell cycle (9) (reviewed by (10)) first pointed to the power of this approach in the field of developmental biology. In this field a particular challenge is to determine how far an embryo, or embryonic region, has advanced along its developmental trajectory: the embryo’s ‘developmental stage’, and in a recent study (11), DNNs were leveraged to classify explanted zebrafish tailbuds into developmental stages (the zebrafish tailbud is a model for posterior spinal cord growth). In contrast to many developing systems, zebrafish embryos can be obtained in large numbers, and in this study, classifiers were trained on primary images of 2 and 3-D datasets.…”

“…In this field, common challenges are to determine how far an embryo has advanced along its developmental trajectory: the embryo's 'developmental stage', or to compare embryos that have been subjected to different experimental regimes. DNNs were recently used to classify the developmental stage of zebrafish tailbuds (an experimental model for posterior spinal cord growth) (7). In this study, networks were trained on 2 and 3-D image datasets, Bayesian optimization was then used to tune network architecture, and success was judged through classification accuracy.…”

Automated staging of zebrafish embryos with deep learning

“…In terms of developmental staging, Pond et al (2021) developed a convolutional neural network (CNN) to stage zebrafish tail-buds at four discrete developmental stages just 1.5 hours apart. Although they trained their network with a small number of images ( < 100) and achieved 100% test accuracy in some cases, the number of test images per class was in several cases as low as 2.…”

Automated staging of zebrafish embryos using machine learning

“…More recently, Guglielmi et al (2021) used an innovative optical projection tomography (OPT) and back-projection technique followed by semi-automated segmentation and quantitation to objectively describe the morphological features of zebrafish embryos in which BMP signaling was perturbed. In terms of developmental staging, Pond et al (2021) recently developed a convolutional neural network (CNN)-based classifier to stage zebrafish tail-buds at four discrete developmental stages, demonstrating that high accuracy can be achieved with small data sets (<100 images). These elegant systems highlight the power of machine learning approaches in the identification of morphological features and discrete developmental stages, but none of these studies extract sufficient information to enable complete temporal developmental profiles to be compared.…”