Pournami P N

Proteins are constructed from amino acid sequences. Their structural classifications include primary, secondary, tertiary, and quaternary, with tertiary and quaternary structures influencing protein function. Because a protein's structure is inextricably connected to its biological function, machine learning algorithms that can better anticipate the structures have the potential to lead to new scientific discoveries in human health and improve our capacity to develop new treatments. Protein secondary structure assignment enriches the structural and functional understanding of proteins. It helps in protein structure comparison and classification studies, besides facilitating secondary and tertiary structure prediction systems. Several secondary structure assignment methods have been developed since the 1980s, most of which are based on hydrogen bond analysis and atomic coordinate features. However, the assignment process becomes complex when protein data includes missing atoms. Deep neural networks are often referred to as universal function approximators because they can approximate any function to produce the desired output when properly designed and trained. Optimised deep learning architectures have already proven their ability to increase performance in a wide range of problems. Recently, the ResNet architecture has garnered significant interest due to its applicability in various areas, including image classification and protein contact map prediction. The proposed model, which is based on the ResNet architecture, assigns secondary structures using Cα atom coordinates. The model achieved an accuracy of 94% when evaluated against the benchmark and independent test sets. The findings encourage the development of new deep learning-based methods that are more generalised across various protein learning tasks. Furthermore, it allows computational biologists to delve deeper into integrating these techniques with experimental methods. The model codes are available at: https://github.com/jisnava/ResNet_for_Structure_Assignments/ .

The past few years have witnessed machine learning techniques take the limelight in multiple research domains. One such domain that has reaped the benefits of machine learning is computer-aided drug discovery, where the search space for candidate drug molecules is decreased using methods such as virtual screening. Current state-of-the-art sequential neural network models have shown promising results and we would like to replicate similar results with virtual screening using the encoded molecular information known as simplified molecular-input line-entry system (SMILES). Our work includes the use of attention-based sequential models — the long short-term memory with attention and an optimized version of the transformer network specifically designed to deal with SMILES (ChemBERTa). We also propose the “Overall Screening Efficacy”, an averaging metric that aggregates and encapsulates the model performance over multiple datasets. We found an overall improvement of about 27% over the benchmark model, which relied on parallelized random forests.

Network-on-Chip (NoC) has become a cost-effective communication interconnect for Tiled Chip Multicore Processor systems. The communication between cores is done through packet exchange. As the computational intensity of applications increases, the amount of packet exchange between cores will also increase. The improper routing of these packets will result in high congestion thereby degrading the system performance. This marks the need for congestion-aware routing in NoC. In the real world, the applications running in NoC create diverse traffic, which in turn creates challenges in routing. Such challenges have resulted in more researchers relying on machine learning algorithms to tackle them. However, the issues pertaining to storage overhead and packet latency prevail in such methodologies. This paper presents an adaptive routing algorithm DeepNR, which uses a deep reinforcement learning approach. The proposed approach considers network information for state representation, routing directions for actions, and queuing delay for reward function. Experiments carried out on synthetic as well as real-time traffics to demonstrate the effectiveness and efficiency of DeepNR using the Gem5 simulator. The results obtained for DeepNR indicate a reduction of up to 21.25% and 44% in overall packet latency under high traffic conditions on real and synthetic traffic respectively, as compared to the existing approaches. Also, DeepNR achieves a throughput of above 90% in both the traffic scenarios.

Mammography has proven to be the most effective tool for detecting breast cancer in its earliest and treatable stage. This paper investigates various phases of mammogram image analysis and different abnormality detection techniques in the mammogram. The present study’s primary aim is to apply the Multifractal theory for the analysis of digital mammograms. The mammogram images are x-raying images of the breast and grayscale in nature. The grayscale mammogram images are processed using image processing techniques and then analyzed by multifractal characteristics. Initially, simple thresholding used to avoid artefacts and noises in the mammograms taken from the MIAS dataset of 322 images. The thresholded images resulted in sharp edges and hence smoothened using a Gaussian filter of appropriate configuration. A novel feature extraction method is proposed based on multifractal spectral parameters. The multifractal characteristics of each mammogram plotted, and multifractal features extracted from the spectrum. The major multifractal features are the bandwidth of the spectrum, the height of the spectrum, maximum and minimum singularity exponent, peak singularity exponent, and the strength of multifractality. Out of the chosen multifractal features, the width of the spectrum, the height of the spectrum, minimum value for the strength of multifractality found to be the most significant features after conducting statistical analysis using ANOVA (Analysis of Variance). We propose a novel lesion localization method based on the extracted multifractal spectral parameters. The pectoral muscle present in the mammogram removed, and the mammogram image divided into four quadrants. The lesion or mass location in a specific quadrant is identified based on the variation in the quadrant’s multifractal characteristics. This method will reduce the search space of lesion in a single mammogram. Thus, the present study revealed that mammogram images exhibit multifractal behaviour. The multifractal parameters appear to be valuable biomarkers for quantitative assessment of breast tissue in the mammogram, which helps to diagnose breast cancer in its early stage.

Antony, JisnaPenikalapati, AkhilReddy, J. Vinod KumarPournami, P. N.Jayaraj, P. B.Determining the native structure of a protein, given its primary sequence is one of the most demanding tasks in computational biology. Traditional protein structure prediction methods are laborious and involve vast conformation search space. Contrarily, deep learning is a rapidly evolving field with outstanding performance at problems where there are complicated relationships between input features and desired outputs. Various deep neural network architectures such as recurrent neural networks, convolution neural networks, deep feed-forward neural networks are becoming popular for solving problems in protein science. This work mainly concentrates on prediction of three-dimensional structure of proteins from the given primary sequences using deep learning techniques. Long short-term memory (LSTM) and bidirectional LSTM (BLSTM) neural network architectures are used for predicting protein tertiary structures from primary sequences. The result shows that single-layer BLSTM networks fed with primary sequence and position-specific scoring matrix data gives better accuracy compared to LSTM and two-layer BLSTM models. This study may get benefited to the computational biologists working in the area of protein structure prediction.

Image registration is an indispensible task required in many image processing applications, which geometrically aligns multiple images of a scene, with differences caused due to time, viewpoint or by heterogeneous sensors. Feature-based registration algorithms are more robust to handle complex geometrical and intensity distortions when compared to area-based techniques. A set of appropriate geometrically invariant features forms the cornerstone for a feature-based registration framework. Feature point or interest point detectors extract salient structures such as points, lines, curves, regions, edges, or objects from the images. A novel interest point detector is presented in this paper. This algorithm computes interest points in a grayscale image by utilizing a graph centrality measure derived from a local image network. This approach exhibits superior repeatability in images where large photometric and geometric variations are present. The practical utility of this highly repeatable feature detector is evident from the simulation results.

Image registration is an elementary task in Computer Vision, which geometrically aligns multiple images of a scene, captured at different times, from various viewpoints, or by heterogeneous sensors. The optimisation strategy we employ for achieving the optimal set of transformation vectors is a major factor that determines the success and effectiveness of an automatic registration procedure. This paper discusses a scheme to modify the conventional Particle Swarm Optimisation (PSO) algorithm for better search space exploration and for faster convergence. While PSO is running, after half of the total number of iterations, find the particle which is in worst position in space, then reposition that particle by mean value of its current position and the global solution. It is observed that re-positioning the worst particle in space helps that particle from premature convergence to a local optimum solution and motivates the particle to generate unique search directions, which increased the possibility of finding the globally best solution. An image registration algorithm using this modified PSO method is also presented. From the experimental results presented here, it is visible that the proposed algorithm guarantees superior results in terms of registration accuracy and reduced execution time, even in the case of large deformations between the reference and float images.