Domain-Specific Pre-training Improves Confidence in Whole Slide Image Classification.

Whole Slide Images (WSIs) or histopathology images are used in digital pathology. WSIs pose great challenges to deep learning models for clinical diagnosis, owing to their size and lack of pixel-level annotations. With the recent advancements in computational pathology, newer multiple-instance learning-based models have been proposed. Multiple-instance learning for WSIs necessitates creating patches and uses the encoding of these patches for diagnosis. These models use generic pre-trained models (ResNet-50 pre-trained on ImageNet) for patch encoding. The recently proposed KimiaNet, a DenseNet121 model pre-trained on TCGA slides, is a domain-specific pre-trained model. This paper shows the effect of domain-specific pre-training on WSI classification. To investigate the effect of domain-specific pre-training, we considered the current state-of-the-art multiple-instance learning models, 1) CLAM, an attention-based model, and 2) TransMIL, a self-attention-based model, and evaluated the models' confidence and predictive performance in detecting primary brain tumors - gliomas. Domain-specific pre-training improves the confidence of the models and also achieves a new state-of-the-art performance of WSI-based glioma subtype classification, showing a high clinical applicability in assisting glioma diagnosis. We will publicly share our code and experimental results at https://github.com/soham-chitnis10/WSI-domain-specific.

An AI-assisted Investigation of Tumor-Associated Macrophages and their Polarization in Colorectal Cancer.

Tumor-associated Macrophages (or TAMs) are amongst the most common cells that play a significant role in the initiation and progression of colorectal cancer (CRC). Recently, Ghosh et al. proposed distinguishing signatures for identifying macrophage polarization states, namely, immuno-reactive and immuno-tolerant, using the concept of Boolean implications and Boolean networks. Their signature, called the Signature of Macrophage Reactivity and Tolerance (SMaRT), comprises of 338 human genes (equivalently, 298 mouse genes). However, SMaRT was constructed using datasets that were not specialized towards any particular disease. In this paper, (a) we perform a comprehensive analysis of the SMaRT signature on single-cell human and mouse colorectal cancer RNA-seq datasets; (b) we then adopt a technique akin to transfer learning to construct a "refined" SMaRT signature for investigating TAMs and their polarization in the CRC tumor microenvironment. Towards validation of our refined gene signature, we use (a) 5 pseudo-bulk RNA-seq datasets derived from single-cell human datasets; and (b) 5 large-cohort microarray datasets from humans. Furthermore, we investigate the translational potential of our refined gene signature in problems related to MSS/MSI (4 datasets) and CIMP+/CIMP- status (4 datasets). Overall, our refined gene signature and its extensive validation provide a path for its adoption in clinical practice in diagnosing colorectal cancer and associated attributes.

AI-assisted Emergency Healthcare using Vehicular Network and Support Vector Machine.

The COVID-19 pandemic has created an emergency across the globe. The number of corona positive and death cases is still rising worldwide. All countries' governments are taking various steps to control the infection of COVID-19. One step to control the coronavirus's spreading is to quarantine. But the number of active cases at the quarantine center is increasing daily. Also, the doctors, nurses, and paramedical staff providing service to the people at the quarantine center are getting infected. This demands the automatic and regular monitoring of people at the quarantine center. This paper proposed a novel and automated method for monitoring people at the quarantine center in two phases. These are the health data transmission phase and health data analysis phase. The health data transmission phase proposed a geographic-based routing that involves components like Network-in-box, Roadside-unit, and vehicles. An effective route is determined using route value to transmit data from the quarantine center to the observation center. The route value depends on the factors such as density, shortest path, delay, vehicular data carrying delay, and attenuation. The performance metrics considered for this phase are E2E delay, number of network gaps, and packet delivery ratio, and the proposed work performs better than the existing routing like geographic source routing, anchor-based street traffic aware routing, Peripheral node based GEographic DIstance Routing . The analysis of health data is done at the observation center. In the health data analysis phase, the health data is classified into multi-class using a support vector machine. There are four categories of health data: normal, low-risk, medium-risk, and high-risk. The parameters used to measure the performance of this phase are precision, recall, accuracy, and F-1 score. The overall testing accuracy is found to be 96.8%, demonstrating strong potential for our technique to be adopted in practice.

A review of some techniques for inclusion of domain-knowledge into deep neural networks.

We present a survey of ways in which existing scientific knowledge are included when constructing models with neural networks. The inclusion of domain-knowledge is of special interest not just to constructing scientific assistants, but also, many other areas that involve understanding data using human-machine collaboration. In many such instances, machine-based model construction may benefit significantly from being provided with human-knowledge of the domain encoded in a sufficiently precise form. This paper examines the inclusion of domain-knowledge by means of changes to: the input, the loss-function, and the architecture of deep networks. The categorisation is for ease of exposition: in practice we expect a combination of such changes will be employed. In each category, we describe techniques that have been shown to yield significant changes in the performance of deep neural networks.

Transformational machine learning: Learning how to learn from many related scientific problems.

Almost all machine learning (ML) is based on representing examples using intrinsic features. When there are multiple related ML problems (tasks), it is possible to transform these features into extrinsic features by first training ML models on other tasks and letting them each make predictions for each example of the new task, yielding a novel representation. We call this transformational ML (TML). TML is very closely related to, and synergistic with, transfer learning, multitask learning, and stacking. TML is applicable to improving any nonlinear ML method. We tested TML using the most important classes of nonlinear ML: random forests, gradient boosting machines, support vector machines, k-nearest neighbors, and neural networks. To ensure the generality and robustness of the evaluation, we utilized thousands of ML problems from three scientific domains: drug design, predicting gene expression, and ML algorithm selection. We found that TML significantly improved the predictive performance of all the ML methods in all the domains (4 to 50% average improvements) and that TML features generally outperformed intrinsic features. Use of TML also enhances scientific understanding through explainable ML. In drug design, we found that TML provided insight into drug target specificity, the relationships between drugs, and the relationships between target proteins. TML leads to an ecosystem-based approach to ML, where new tasks, examples, predictions, and so on synergistically interact to improve performance. To contribute to this ecosystem, all our data, code, and our ∼50,000 ML models have been fully annotated with metadata, linked, and openly published using Findability, Accessibility, Interoperability, and Reusability principles (∼100 Gbytes).

Sequence-based discrimination of protein-RNA interacting residues using a probabilistic approach.

Protein interactions with ribonucleic acids (RNA) are well-known to be crucial for a wide range of cellular processes such as transcriptional regulation, protein synthesis or translation, and post-translational modifications. Identification of the RNA-interacting residues can provide insights into these processes and aid in relevant biotechnological manipulations. Owing to their eventual potential in combating diseases and industrial production, several computational attempts have been made over years using sequence- and structure-based information. Recent comparative studies suggest that despite these developments, many problems are faced with respect to the usability, prerequisites, and accessibility of various tools, thereby calling for an alternative approach and perspective supplementation in the prediction scenario. With this motivation, in this paper, we propose the use of a simple-yet-efficient conditional probabilistic approach based on the application of local occurrence of amino acids in the interacting region in a non-numeric sequence feature space, for discriminating between RNA interacting and non-interacting residues. The proposed method has been meticulously tested for robustness using a cross-estimation method showing MCC of 0.341 and F- measure of 66.84%. Upon exploring large scale applications using benchmark datasets available to date, this approach showed an encouraging performance comparable with the state-of-art. The software is available at https://github.com/ABCgrp/DORAEMON.

Gradient gravitational search: An efficient metaheuristic algorithm for global optimization.

The adaptation of novel techniques developed in the field of computational chemistry to solve the concerned problems for large and flexible molecules is taking the center stage with regard to efficient algorithm, computational cost and accuracy. In this article, the gradient-based gravitational search (GGS) algorithm, using analytical gradients for a fast minimization to the next local minimum has been reported. Its efficiency as metaheuristic approach has also been compared with Gradient Tabu Search and others like: Gravitational Search, Cuckoo Search, and Back Tracking Search algorithms for global optimization. Moreover, the GGS approach has also been applied to computational chemistry problems for finding the minimal value potential energy of two-dimensional and three-dimensional off-lattice protein models. The simulation results reveal the relative stability and physical accuracy of protein models with efficient computational cost.