Evaluation of machine learning models for automatic detection of DNA double strand breaks after irradiation using a γH2AX foci assay

Ex vivo radiation sensitivity assessment for individual head and neck cancer patients using deep learning-based automated nuclei and DNA damage foci detection

Introduction

Tumor biopsy tissue response to ex vivo irradiation is potentially an interesting biomarker for in vivo tumor response, therefore, for treatment personalization. Tumor response ex vivo can be characterized by DNA damage response, expressed by the large-scale presence of DNA damage foci in tumor nuclei. Currently, characterizing tumor nuclei and DNA damage foci is a manual process that takes hours per patient and is subjective to inter-observer variability, which is not feasible in for clinical decision making. Therefore, our goal was to develop a method to automatically segment nuclei and DNA damage foci in tumor tissue samples treated with radiation ex vivo to characterize the DNA damage response, as potential biomarker for in vivo radio-sensitivity.

Methods

Oral cavity tumor tissue of 21 patients was irradiated ex vivo (5 or 0 Gy), fixated 2 h post-radiation, and used to develop our method for automated nuclei and 53BP1 foci segmentation. The segmentation model used both deep learning and conventional image-analysis techniques. The training (22 %), validation (22 %), and test set (56 %) consisted of thousands of manually segmented nuclei and foci. The segmentations and number of foci per nucleus in the test set were compared to their ground truths.

Results

The automatic nuclei and foci segmentations were highly accurate (Dice = 0.901 and Dice = 0.749, respectively). An excellent correlation (R2 = 0.802) was observed for the foci per nucleus that outperformed reported inter-observation variation. The analysis took ∼ 8 s per image.

Conclusion

This model can replace manual foci analysis for ex vivo irradiation of head-and-neck squamous cell carcinoma tissue, reduces the image-analysis time from hours to minutes, avoids the problem of inter-observer variability, enables assessment of multiple images or conditions, and provides additional information about the foci size. Thereby, it allows for reliable and rapid ex vivo radio-sensitivity assessment, as potential biomarker for response in vivo and treatment personalization.

Multi-scale cellular imaging of DNA double strand break repair

Live-cell and high-resolution fluorescence microscopy are powerful tools to study the organization and dynamics of DNA double-strand break repair foci and specific repair proteins in single cells. This requires specific induction of DNA double-strand breaks and fluorescent markers to follow the DNA lesions in living cells. In this review, where we focused on mammalian cell studies, we discuss different methods to induce DNA double-strand breaks, how to visualize and quantify repair foci in living cells., We describe different (live-cell) imaging modalities that can reveal details of the DNA double-strand break repair process across multiple time and spatial scales. In addition, recent developments are discussed in super-resolution imaging and single-molecule tracking, and how these technologies can be applied to elucidate details on structural compositions or dynamics of DNA double-strand break repair.

HARLEY mitigates user bias and facilitates efficient quantification and co-localization analyses of foci in yeast fluorescence images

Quantification of cellular structures in fluorescence microscopy data is a key means of understanding cellular function. Unfortunately, numerous cellular structures present unique challenges in their ability to be unbiasedly and accurately detected and quantified. In our studies on stress granules in yeast, users displayed a striking variation of up to 3.7-fold in foci calls and were only able to replicate their results with 62-78% accuracy, when re-quantifying the same images. To facilitate consistent results we developed HARLEY (Human Augmented Recognition of LLPS Ensembles in Yeast), a customizable software for detection and quantification of stress granules in S. cerevisiae. After a brief model training on ~ 20 cells the detection and quantification of foci is fully automated and based on closed loops in intensity contours, constrained only by the a priori known size of the features of interest. Since no shape is implied, this method is not limited to round features, as is often the case with other algorithms. Candidate features are annotated with a set of geometrical and intensity-based properties to train a kernel Support Vector Machine to recognize features of interest. The trained classifier is then used to create consistent results across datasets. For less ambiguous foci datasets, a parametric selection is available. HARLEY is an intuitive tool aimed at yeast microscopy users without much technical expertise. It allows batch processing of foci detection and quantification, and the ability to run various geometry-based and pixel-based colocalization analyses to uncover trends or correlations in foci-related data. HARLEY is open source and can be downloaded from https://github.com/lnilya/harley .

A deep learning model (FociRad) for automated detection of γ-H2AX foci and radiation dose estimation

DNA double-strand breaks (DSBs) are the most lethal form of damage to cells from irradiation. γ-H2AX (phosphorylated form of H2AX histone variant) has become one of the most reliable and sensitive biomarkers of DNA DSBs. However, the γ-H2AX foci assay still has limitations in the time consumed for manual scoring and possible variability between scorers. This study proposed a novel automated foci scoring method using a deep convolutional neural network based on a You-Only-Look-Once (YOLO) algorithm to quantify γ-H2AX foci in peripheral blood samples. FociRad, a two-stage deep learning approach, consisted of mononuclear cell (MNC) and γ-H2AX foci detections. Whole blood samples were irradiated with X-rays from a 6 MV linear accelerator at 1, 2, 4 or 6 Gy. Images were captured using confocal microscopy. Then, dose-response calibration curves were established and implemented with unseen dataset. The results of the FociRad model were comparable with manual scoring. MNC detection yielded 96.6% accuracy, 96.7% sensitivity and 96.5% specificity. γ-H2AX foci detection showed very good F1 scores (> 0.9). Implementation of calibration curve in the range of 0-4 Gy gave mean absolute difference of estimated doses less than 1 Gy compared to actual doses. In addition, the evaluation times of FociRad were very short (< 0.5 min per 100 images), while the time for manual scoring increased with the number of foci. In conclusion, FociRad was the first automated foci scoring method to use a YOLO algorithm with high detection performance and fast evaluation time, which opens the door for large-scale applications in radiation triage.

Gray-Level Co-occurrence Matrix Analysis for the Detection of Discrete, Ethanol-Induced, Structural Changes in Cell Nuclei: An Artificial Intelligence Approach

Gray-level co-occurrence matrix (GLCM) analysis is a contemporary and innovative computational method for the assessment of textural patterns, applicable in almost any area of microscopy. The aim of our research was to perform the GLCM analysis of cell nuclei in Saccharomyces cerevisiae yeast cells after the induction of sublethal cell damage with ethyl alcohol, and to evaluate the performance of various machine learning (ML) models regarding their ability to separate damaged from intact cells. For each cell nucleus, five GLCM parameters were calculated: angular second moment, inverse difference moment, GLCM contrast, GLCM correlation, and textural variance. Based on the obtained GLCM data, we applied three ML approaches: neural network, random trees, and binomial logistic regression. Statistically significant differences in GLCM features were observed between treated and untreated cells. The multilayer perceptron neural network had the highest classification accuracy. The model also showed a relatively high level of sensitivity and specificity, as well as an excellent discriminatory power in the separation of treated from untreated cells. To the best of our knowledge, this is the first study to demonstrate that it is possible to create a relatively sensitive GLCM-based ML model for the detection of alcohol-induced damage in Saccharomyces cerevisiae cell nuclei.

Assessment of Neuronal Damage in Brain Slice Cultures Using Machine Learning Based on Spatial Features

Neuronal damage presents a major health issue necessitating extensive research to identify mechanisms of neuronal cell death and potential therapeutic targets. Commonly used models are slice cultures out of different brain regions extracted from mice or rats, excitotoxically, ischemic, or traumatically lesioned and subsequently treated with potential neuroprotective agents. Thereby cell death is regularly assessed by measuring the propidium iodide (PI) uptake or counting of PI-positive nuclei. The applied methods have a limited applicability, either in terms of objectivity and time consumption or regarding its applicability. Consequently, new tools for analysis are needed. Here, we present a framework to mimic manual counting using machine learning algorithms as tools for semantic segmentation of PI-positive dead cells in hippocampal slice cultures. Therefore, we trained a support vector machine (SVM) to classify images into either “high” or “low” neuronal damage and used naïve Bayes, discriminant analysis, random forest, and a multilayer perceptron (MLP) as classifiers for segmentation of dead cells. In our final models, pixel-wise accuracies of up to 0.97 were achieved using the MLP classifier. Furthermore, a SVM-based post-processing step was introduced to differentiate between false-positive and false-negative detections using morphological features. As only very few false-positive objects and thus training data remained when using the final model, this approach only mildly improved the results. A final object splitting step using Hough transformations was used to account for overlap, leading to a recall of up to 97.6% of the manually assigned PI-positive dead cells. Taken together, we present an analysis tool that can help to objectively and reproducibly analyze neuronal damage in brain-derived slice cultures, taking advantage of the morphology of pycnotic cells for segmentation, object splitting, and identification of false positives.

Deoxycholic acid-modified microporous SiO2nanofibers mimicking colorectal microenvironment to optimize radiotherapy-chemotherapy combined therapy

Radiotherapy and chemotherapy remain the main therapeutics for colorectal cancer. However, due to their inevitable side effects on nomal tissues, it is necessary to evaluate the toxicity of radio-/chemotherapy regimens. The newly developedin vitrohigh throughput strategy is promising for these assessments. Nevertheless, the currently monolayer culture condition adopted in the preclinical screening processesin vitrohas been proved not so efficient asin vivosince its poor physiological similarity toin vivomicroenvironment. Herein, we fabricated microporous SiO₂nanofiber mats and further bioactivated with deoxycholic acid (DCA) to mimic the chemical signals in the colorectal cancer microenvironment forin vitroregimen assessment of radiotherapy and chemotherapy. The colorectal cancer cells contacted with the DCA-modified SiO₂nanofiber (SiO₂-DCA NF) mats spatially, and the human intestinal epithelial cell on SiO₂-DCA NF mats exhibited better x-ray and cisplatin tolerance. The distinguishable irradiation and drug tolerance of cells on SiO₂-DCA NF mats indicated that the actual microenvironment of intestine might instruct colorectal cancer differently compared with the common biological experiments. The presented DCA-modified microporous SiO₂nanofibrous mats endowing a better mimicry of colorectal micro-environment, would provide a promising platform forin vitroassessment of radio-/chemotherapy regimens.

Application of artificial intelligence for detection of chemico-biological interactions associated with oxidative stress and DNA damage

In recent years, various AI-based methods have been developed in order to uncover chemico-biological interactions associated with DNA damage and oxidative stress. Various decision trees, bayesian networks, random forests, logistic regression models, support vector machines as well as deep learning tools, have great potential in the area of molecular biology and toxicology, and it is estimated that in the future, they will greatly contribute to our understanding of molecular and cellular mechanisms associated with DNA damage and repair. In this concise review, we discuss recent attempts to build machine learning tools for assessment of radiation - induced DNA damage as well as algorithms that can analyze the data from the most frequently used DNA damage assays in molecular biology. We also review recent works on the detection of antioxidant proteins with machine learning, and the use of AI-related methods for prediction and evaluation of noncoding DNA sequences. Finally, we discuss previously published research on the potential application of machine learning tools in aging research.