Tyler O'Neal, Staff Editor GOVERNMENT

Dartmouth team develops new machine learning approach that detects esophageal cancer better than current methods

Researchers at Dartmouth's Norris Cotton Cancer Center have created a deep learning model that can accurately identify cancerous esophagus tissue on microscopy images without the time-consuming manual data input required for current methods

Recently, deep learning methods have shown promising results for analyzing histological patterns in microscopy images. These approaches, however, require a laborious, high-cost, manual annotation process by pathologists called "region-of-interest annotations." A research team at Dartmouth and Dartmouth-Hitchcock Norris Cotton Cancer Center, led by Saeed Hassanpour, Ph.D., has addressed this shortcoming of current methods by developing a novel attention-based deep learning method that automatically learns clinically important regions on whole-slide images to classify them.

The team tested their new approach for identifying cancerous and precancerous esophagus tissue on high-resolution microscopy images without training on region-of-interest annotations. "Our new approach outperformed the current state-of-the-art approach that requires these detailed annotations for its training," concludes Hassanpour. Their results, "Detection of Cancerous and Precancerous Esophagus Tissue on Histopathology Slides Using Attention-Based Deep Neural Networks" will publish in JAMA Network Open in early November 2019. {module INSIDE STORY}

For histopathology image analysis, the manual annotation process typically outlines the regions of interest on a high-resolution whole slide image to facilitate training the computer model. "Data annotation is the most time-consuming and laborious bottleneck in developing modern deep learning methods," notes Hassanpour. "Our study shows that deep learning models for histopathology slides analysis can be trained with labels only at the tissue level, thus removing the need for high-cost data annotation and creating new opportunities for expanding the application of deep learning in digital pathology."

The team proposed the network for Barrett esophagus and esophageal adenocarcinoma detection and found that its performance exceeds that of the existing state-of-the-art method. "The result is significant because our method is based solely on tissue-level annotations, unlike existing methods that are based on manually annotated regions," says Hassanpour. He expects this model to open new avenues for applying deep learning to digital pathology. "Our method would facilitate a more extensive range of research on analyzing histopathology images that were previously not possible due to the lack of detailed annotations. Clinical deployment of such systems could assist pathologists in reading histopathology slides more accurately and efficiently, which is a critical task for the cancer diagnosis, predicting prognosis, and treatment of cancer patients."

Looking ahead, Hassanpour's team is planning to validate their model further by testing it on data from other institutions and running prospective clinical trials. They also plan to apply the proposed model to histological images of other types of tumors and lesions for which training data are scarce or bounding box annotations are not available.

UK researchers produce a human protein co-regulation map that reveals new insights into protein functions

Tyler O'Neal, Staff Editor GOVERNMENT

Proteins are key molecules in living cells. They are responsible for nearly every task of cellular life and are essential for the maintenance of the structure, function, and regulation of tissues and organs in the human body. The cells in the human body can form thousands of different types of proteins (the so-called proteome), which perform a plethora of diverse functions, all crucial for cell viability and human health. Assigning functions to the vast array of proteins present in our cells remains a challenging task in cell biology.

Scientists have now produced a co-regulation map of the human proteome, which was able to capture relationships between proteins that do not physically interact or co-localize. This will enable the prediction and assignment of functions to uncharacterized human proteins. The co-regulation map can be explored at http://www.proteomeHD.net. CAPTION Co-regulation map shows associations between human proteins. CREDIT Michael Schrader - University of Exeter{module In article}

Many of the functions of human proteins are still unknown, but researchers at the Wellcome Centre for Cell Biology, the University of Edinburgh and the Institute of Biotechnology, Technische Universität Berlin have applied large scale quantitative proteomics and machine-learning to produce a protein covariation dataset of the human proteome. The dataset forms the basis of a co-regulation map which can be used to predict the potential function of uncharacterized human proteins.

"In this study, we took thousands of mass spectrometry experiments that other laboratories had published over the past few years and re-purposed them in a way that was completely unrelated to what the original authors had intended. We used a machine-learning algorithm to mine this huge collection of data and managed to assign a biological function to hundreds of proteins (genes) that were previously uncharacterized", said Professor Juri Rappsilber and Dr. Georg Kustatscher, of the University of Edinburgh and TU Berlin.

Exploring the map revealed unexpected co-regulation partners, namely the peroxisomal membrane protein PEX11β with mitochondrial respiration factors. In cooperation with Professor Michael Schrader and his team at the University of Exeter, this led to the identification of a novel interaction between two crucial cellular organelles - mitochondria and peroxisomes.

"Peroxisomes and mitochondria in mammals are intimately linked and cooperate in the breakdown of fatty acids and cellular energy balance. Using live-cell imaging we revealed that PEX11β mediates the formation of membrane protrusions, which emanate from peroxisomes and interact with mitochondria. They likely function in the metabolic cooperation and crosstalk between both organelles, and may facilitate the transfer of metabolites during mitochondrial energy (ATP) production", said Professor Michael Schrader, of the University of Exeter.

The scientists at the University of Edinburgh created the website http://www.proteomeHD.net to allow users to search for a protein of interest, showing its position in the co-regulation map together with any co-regulation partners.

The online map is interactive and zoomable, making it easy to explore the neighborhood of a query protein. It is designed to support researchers in exploring co-regulation data at multiple scales, to validate existing hypotheses or to create new ones.

"With an ever-increasing amount of protein expression data being made available, protein co-expression analysis has huge potential for gene function annotation. In a time when "big data" becomes more and more relevant for life science, a key lessons we learned from this project is: never throw away your data - they can be re-purposed, recycled, and with the right tools there is plenty more information and knowledge that can be extracted from them", said Dr. Georg Kustatscher, of the University of Edinburgh.

HIV spreads through direct cell-to-cell contact

Tyler O'Neal, Staff Editor GOVERNMENT

CAPTION Microscopic recording and supercomputer model of the interaction between infected cells (green) and non-infected cells (red) in collagen structures (grey). CREDIT Oliver Fackler / Frederik Graw

German researchers investigate infection dynamics in tissue-like three-dimensional cell cultures

The spread of pathogens like the human immunodeficiency virus (HIV) is often studied in a test tube, i.e. in two-dimensional cell cultures, even though it hardly reflects the much more complex conditions in the human body. Using innovative cell culture systems, quantitative image analysis, and supercomputer simulations, an interdisciplinary team of scientists from Heidelberg University has now explored how HIV spreads in three-dimensional tissue-like environments. The researchers' results show that the tissue structure forces the virus to spread through direct cell-to-cell contact. 

Despite over 30 years of research, many key aspects of how HIV, the causative agent of the acquired immune deficiency syndrome (AIDS) spreads are still not understood. One of these unresolved questions concerns the interactions between the virus with the environment in the human body. Traditionally it has been assumed that infected cells release viral particles which then diffuse and eventually infect other cells. But it is also possible that viral particles are directly transferred from one infected cell to the next through close contact. Until now it was unknown which of these modes of transmission prevailed in tissue. "Studies on HIV replication in the lab are mostly conducted in simple cell culture experiments in plastic dishes that do not reflect the complex architecture and heterogeneity of tissue", explains study director Prof. Dr Oliver Fackler of the Center for Integrative Infectious Disease Research (CIID) at Heidelberg University Hospital. {module In-article}

In their approach, the Heidelberg researchers took into account that the so-called CD4 T helper cells, the preferred cell type infected by HIV, are highly motile in their physiological environment. They used a novel cell culture system, in which a three-dimensional scaffold was generated with the help of collagen. This allowed for maintaining the cells' mobility and monitoring primary CD4 T cells infected with HIV-1 in a tissue-like environment over the course of several weeks. Using this innovative approach, the researchers measured a number of factors that characterise cell motility, virus replication, and the gradual loss of CD4 T helper cells. "This yielded a very complex set of data that was impossible to interpret without the help from scientists of other disciplines", explains Dr Andrea Imle, who worked on the project during her PhD at the CIID.

In analysing the data, the scientists who conducted the experiments collaborated with colleagues from the fields of image processing, theoretical biophysics and mathematical modelling. Together they were able to characterise the complex behaviour of cells and viruses and simulate it on the computer. This made it possible to make important predictions on the key processes that determine HIV-1 spread in these 3D cultures, which were confirmed by subsequent experimentation. "Our interdisciplinary study is a good example of how iterative cycles of experimentation and simulation can help to quantitatively analyse a complex biological process", states Prof. Dr Ulrich Schwarz of the Institute for Theoretical Physics at Heidelberg University.

The data analysis revealed that the 3D environment of the cell culture system suppresses infection with a cell-free virus while simultaneously promoting direct virus transmission from cell to cell. "Our models allowed us to integrate short single-cell microscopy films with long-term cell population measurements and thereby to estimate the minimal time span required for cell-to-cell contacts to transmit infection", explains Dr Frederik Graw of the BioQuant Centre of Heidelberg University. The researchers hope that these findings will eventually lead to new therapeutic approaches in the treatment of HIV.