Jagannathan Ramanujam

SynerGNet is a novel approach to predicting drug synergy against cancer cell lines. In this study, we discuss in detail the construction process of SynerGNet, emphasizing its comprehensive design tailored to handle complex data patterns. Additionally, we investigate a counterintuitive phenomenon when integrating more augmented data into the training set results in an increase in testing loss alongside improved predictive accuracy. This sheds light on the nuanced dynamics of model learning. Further, we demonstrate the effectiveness of strong regularization techniques in mitigating overfitting, ensuring the robustness and generalization ability of SynerGNet. Finally, the continuous performance enhancements achieved through the integration of augmented data are highlighted. By gradually increasing the amount of augmented data in the training set, we observe substantial improvements in model performance. For instance, compared to models trained exclusively on the original data, the integration of the augmented data can lead to a 5.5% increase in the balanced accuracy and a 7.8% decrease in the false positive rate. Through rigorous benchmarks and analyses, our study contributes valuable insights into the development and optimization of predictive models in biomedical research.

Semantic segmentation is a core computer vision problem, but the high costs of data annotation have hindered its wide application. Weakly-Supervised Semantic Segmentation (WSSS) offers a cost-efficient workaround to extensive labeling in comparison to fully-supervised methods by using partial or incomplete labels. Existing WSSS methods have difficulties in learning the boundaries of objects leading to poor segmentation results. We propose a novel and effective framework that addresses these issues by leveraging visual foundation models inside the bounding box. Adopting a two-stage WSSS framework, our proposed network consists of a pseudo-label generation module and a segmentation module. The first stage leverages Segment Anything Model (SAM) to generate high-quality pseudo-labels. To alleviate the problem of delineating precise boundaries, we adopt SAM inside the bounding box with the help of another pre-trained foundation model (e.g., Grounding-DINO). Furthermore, we eliminate the necessity of using the supervision of image labels, by employing CLIP in classification. Then in the second stage, the generated high-quality pseudo-labels are used to train an off-the-shelf segmenter that achieves the state-of-the-art performance on PASCAL VOC 2012 and MS COCO 2014.

This paper provides a review on representation learning for videos. We classify recent spatiotemporal feature learning methods for sequential visual data and compare their pros and cons for general video analysis. Building effective features for videos is a fundamental problem in computer vision tasks involving video analysis and understanding. Existing features can be generally categorized into spatial and temporal features. Their effectiveness under variations of illumination, occlusion, view and background are discussed. Finally, we discuss the remaining challenges in existing deep video representation learning studies.

Drug combination therapy shows promise in cancer treatment by addressing drug resistance, reducing toxicity, and enhancing therapeutic efficacy. However, the intricate and dynamic nature of biological systems makes identifying potential synergistic drugs a costly and time-consuming endeavor. To facilitate the development of combination therapy, techniques employing artificial intelligence have emerged as a transformative solution, providing a sophisticated avenue for advancing existing therapeutic approaches. In this study, we developed SynerGNet, a graph neural network model designed to accurately predict the synergistic effect of drug pairs against cancer cell lines. SynerGNet utilizes cancer-specific featured graphs created by integrating heterogeneous biological features into the human protein–protein interaction network, followed by a reduction process to enhance topological diversity. Leveraging synergy data provided by AZ-DREAM Challenges, the model yields a balanced accuracy of 0.68, significantly outperforming traditional machine learning. Encouragingly, augmenting the training data with carefully constructed synthetic instances improved the balanced accuracy of SynerGNet to 0.73. Finally, the results of an independent validation conducted against DrugCombDB demonstrated that it exhibits a strong performance when applied to unseen data. SynerGNet shows a great potential in detecting drug synergy, positioning itself as a valuable tool that could contribute to the advancement of combination therapy for cancer treatment.

Deregulated protein kinases are crucial in promoting cancer cell proliferation and driving malignant cell signaling. Although these kinases are essential targets for cancer therapy due to their involvement in cell development and proliferation, only a small part of the human kinome has been targeted by drugs. A comprehensive scoring system is needed to evaluate and prioritize clinically relevant kinases. We recently developed CancerOmicsNet, an artificial intelligence model employing graph-based algorithms to predict the cancer cell response to treatment with kinase inhibitors. The performance of this approach has been evaluated in large-scale benchmarking calculations, followed by the experimental validation of selected predictions against several cancer types. To shed light on the decision-making process of CancerOmicsNet and to better understand the role of each kinase in the model, we employed a customized saliency map with adjustable channel weights. The saliency map, functioning as an explainable AI tool, allows for the analysis of input contributions to the output of a trained deep-learning model and facilitates the identification of essential kinases involved in tumor progression. The comprehensive survey of biomedical literature for essential kinases selected by CancerOmicsNet demonstrated that it could help pinpoint potential druggable targets for further investigation in diverse cancer types.

The binding of small organic molecules to protein targets is fundamental to a wide array of cellular functions. It is also routinely exploited to develop new therapeutic strategies against a variety of diseases. On that account, the ability to effectively detect and classify ligand binding sites in proteins is of paramount importance to modern structure-based drug discovery. These complex and non-trivial tasks require sophisticated algorithms from the field of artificial intelligence to achieve a high prediction accuracy. In this communication, we describe GraphSite, a deep learning-based method utilizing a graph representation of local protein structures and a state-of-the-art graph neural network to classify ligand binding sites. Using neural weighted message passing layers to effectively capture the structural, physicochemical, and evolutionary characteristics of binding pockets mitigates model overfitting and improves the classification accuracy. Indeed, comprehensive cross-validation benchmarks against a large dataset of binding pockets belonging to 14 diverse functional classes demonstrate that GraphSite yields the class-weighted F1-score of 81.7%, outperforming other approaches such as molecular docking and binding site matching. Further, it also generalizes well to unseen data with the F1-score of 70.7%, which is the expected performance in real-world applications. We also discuss new directions to improve and extend GraphSite in the future.

Spatial reasoning is a critical skill in many everyday tasks and in science, technology, engineering, and mathematics disciplines. The current study examined how training on mental rotation (a spatial reasoning task) impacts the completeness of an encoded representation and the ability to rotate the representation. We used a multisession, multimethod design with an active control group to determine how mental rotation ability impacts performance for a trained stimulus category and an untrained stimulus category. Participants in the experimental group (n = 18) showed greater improvement than the active control group (n = 18) on the mental rotation tasks. The number of saccades between objects decreased and saccade amplitude increased after training, suggesting that participants in the experimental group encoded more of the object and possibly had more complete mental representations after training. Functional magnetic resonance imaging data revealed distinct neural activation associated with mental rotation, notably in the right motor cortex and right lateral occipital cortex. These brain areas are often associated with rotation and encoding complete representations, respectively. Furthermore, logistic regression revealed that activation in these brain regions during the post-training scan significantly predicted training group assignment. Overall, the current study suggests that effective mental rotation training protocols should aim to improve the encoding and manipulation of mental representations.

Fast and accurate classification of ligand-binding sites in proteins with respect to the class of binding molecules is invaluable not only to the automatic functional annotation of large datasets of protein structures but also to projects in protein evolution, protein engineering and drug development. Deep learning techniques, which have already been successfully applied to address challenging problems across various fields, are inherently suitable to classify ligand-binding pockets. Our goal is to demonstrate that off-the-shelf deep learning models can be employed with minimum development effort to recognize nucleotide- and heme-binding sites with a comparable accuracy to highly specialized, voxel-based methods. We developed BionoiNet, a new deep learning-based framework implementing a popular ResNet model for image classification. BionoiNet first transforms the molecular structures of ligand-binding sites to 2D Voronoi diagrams, which are then used as the input to a pretrained convolutional neural network classifier. The ResNet model generalizes well to unseen data achieving the accuracy of 85.6% for nucleotide- and 91.3% for heme-binding pockets. BionoiNet also computes significance scores of pocket atoms, called BionoiScores, to provide meaningful insights into their interactions with ligand molecules. BionoiNet is a lightweight alternative to computationally expensive 3D architectures. BionoiNet is implemented in Python with the source code freely available at: https://github.com/CSBG-LSU/BionoiNet. Supplementary data are available at Bioinformatics online.

Sparse tiling is a technique to fuse loops that access common data, thus increasing data locality. Unlike traditional loop fusion or blocking, the loops may have different iteration spaces and access shared datasets through indirect memory accesses, such as A[map[i]]—hence the name “sparse.” One notable example of such loops arises in discontinuous-Galerkin finite element methods, because of the computation of numerical integrals over different domains (e.g., cells, facets). The major challenge with sparse tiling is implementation—not only is it cumbersome to understand and synthesize, but it is also onerous to maintain and generalize, as it requires a complete rewrite of the bulk of the numerical computation. In this article, we propose an approach to extend the applicability of sparse tiling based on raising the level of abstraction. Through a sequence of compiler passes, the mathematical specification of a problem is progressively lowered, and eventually sparse-tiled C for-loops are generated. Besides automation, we advance the state-of-the-art by introducing a revisited, more efficient sparse tiling algorithm; support for distributed-memory parallelism; a range of fine-grained optimizations for increased runtime performance; implementation in a publicly available library, SLOPE; and an in-depth study of the performance impact in Seigen, a real-world elastic wave equation solver for seismological problems, which shows speed-ups up to 1.28× on a platform consisting of 896 Intel Broadwell cores.