Fine-tuning TrailMap: The utility of transfer learning to improve the performance of deep learning in axon segmentation of light-sheet microscopy images.

Light-sheet microscopy has made possible the 3D imaging of both fixed and live biological tissue, with samples as large as the entire mouse brain. However, segmentation and quantification of that data remains a time-consuming manual undertaking. Machine learning methods promise the possibility of automating this process. This study seeks to advance the performance of prior models through optimizing transfer learning. We fine-tuned the existing TrailMap model using expert-labeled data from noradrenergic axonal structures in the mouse brain. By changing the cross-entropy weights and using augmentation, we demonstrate a generally improved adjusted F1-score over using the originally trained TrailMap model within our test datasets.

Fine-tuning TrailMap: The utility of transfer learning to improve the performance of deep learning in axon segmentation of light-sheet microscopy images.

Light-sheet microscopy has made possible the 3D imaging of both fixed and live biological tissue, with samples as large as the entire mouse brain. However, segmentation and quantification of that data remains a time-consuming manual undertaking. Machine learning methods promise the possibility of automating this process. This study seeks to advance the performance of prior models through optimizing transfer learning. We fine-tuned the existing TrailMap model using expert-labeled data from noradrenergic axonal structures in the mouse brain. By fine-tuning the final two layers of the neural network at a lower learning rate of the TrailMap model, we demonstrate an improved recall and an occasionally improved adjusted F1-score within our test dataset over using the originally trained TrailMap model.

Evaluating Stage Motion for Automated Electron Microscopy.

Precise control is an essential and elusive quality of emerging self-driving transmission electron microscopes (TEMs). It is widely understood these instruments must be capable of performing rapid, high-volume, and arbitrary movements for practical self-driving operation. However, stage movements are difficult to automate at scale, owing to mechanical instability, hysteresis, and thermal drift. Such difficulties pose major barriers to artificial intelligence-directed microscope designs that require repeatable, precise movements. To guide design of emerging instruments, it is necessary to understand the behavior of existing mechanisms to identify rate limiting steps for full autonomy. Here, we describe a general framework to evaluate stage motion in any TEM. We define metrics to evaluate stage degrees of freedom, propose solutions to improve performance, and comment on fundamental limits to automated experimentation using present hardware.

Classifying metal-binding sites with neural networks.

To advance our ability to predict impacts of the protein scaffold on catalysis, robust classification schemes to define features of proteins that will influence reactivity are needed. One of these features is a protein's metal-binding ability, as metals are critical to catalytic conversion by metalloenzymes. As a step toward realizing this goal, we used convolutional neural networks (CNNs) to enable the classification of a metal cofactor binding pocket within a protein scaffold. CNNs enable images to be classified based on multiple levels of detail in the image, from edges and corners to entire objects, and can provide rapid classification. First, six CNN models were fine-tuned to classify the 20 standard amino acids to choose a performant model for amino acid classification. This model was then trained in two parallel efforts: to classify a 2D image of the environment within a given radius of the central metal binding site, either an Fe ion or a [2Fe-2S] cofactor, with the metal visible (effort 1) or the metal hidden (effort 2). We further used two sub-classifications of the [2Fe-2S] cofactor: (1) a standard [2Fe-2S] cofactor and (2) a Rieske [2Fe-2S] cofactor. The accuracy for the model correctly identifying all three defined features was >95%, despite our perception of the increased challenge of the metalloenzyme identification. This demonstrates that machine learning methodology to classify and distinguish similar metal-binding sites, even in the absence of a visible cofactor, is indeed possible and offers an additional tool for metal-binding site identification in proteins.

An Automated Scanning Transmission Electron Microscope Guided by Sparse Data Analytics.

Artificial intelligence (AI) promises to reshape scientific inquiry and enable breakthrough discoveries in areas such as energy storage, quantum computing, and biomedicine. Scanning transmission electron microscopy (STEM), a cornerstone of the study of chemical and materials systems, stands to benefit greatly from AI-driven automation. However, present barriers to low-level instrument control, as well as generalizable and interpretable feature detection, make truly automated microscopy impractical. Here, we discuss the design of a closed-loop instrument control platform guided by emerging sparse data analytics. We hypothesize that a centralized controller, informed by machine learning combining limited a priori knowledge and task-based discrimination, could drive on-the-fly experimental decision-making. This platform may unlock practical, automated analysis of a variety of material features, enabling new high-throughput and statistical studies.

Integration of Infant Metabolite, Genetic, and Islet Autoimmunity Signatures to Predict Type 1 Diabetes by Age 6 Years.

CONTEXT: Biomarkers that can accurately predict risk of type 1 diabetes (T1D) in genetically predisposed children can facilitate interventions to delay or prevent the disease. OBJECTIVE: This work aimed to determine if a combination of genetic, immunologic, and metabolic features, measured at infancy, can be used to predict the likelihood that a child will develop T1D by age 6 years. METHODS: Newborns with human leukocyte antigen (HLA) typing were enrolled in the prospective birth cohort of The Environmental Determinants of Diabetes in the Young (TEDDY). TEDDY ascertained children in Finland, Germany, Sweden, and the United States. TEDDY children were either from the general population or from families with T1D with an HLA genotype associated with T1D specific to TEDDY eligibility criteria. From the TEDDY cohort there were 702 children will all data sources measured at ages 3, 6, and 9 months, 11.4% of whom progressed to T1D by age 6 years. The main outcome measure was a diagnosis of T1D as diagnosed by American Diabetes Association criteria. RESULTS: Machine learning-based feature selection yielded classifiers based on disparate demographic, immunologic, genetic, and metabolite features. The accuracy of the model using all available data evaluated by the area under a receiver operating characteristic curve is 0.84. Reducing to only 3- and 9-month measurements did not reduce the area under the curve significantly. Metabolomics had the largest value when evaluating the accuracy at a low false-positive rate. CONCLUSION: The metabolite features identified as important for progression to T1D by age 6 years point to altered sugar metabolism in infancy. Integrating this information with classic risk factors improves prediction of the progression to T1D in early childhood.