Biology subjects

Chourasia, P.

Publications and source records attributed to Chourasia, P..

6 recordsLinked to original sources

bioRxiv2024

Preserving Hidden Hierarchical Structure: Poincare Distance for Enhanced Genomic Sequence Analysis

The analysis of large volumes of molecular (genomic, proteomic, etc.) sequences has become a significant research field, especially after the recent coronavirus pandemic. Although it has proven beneficial to sequence analysis, machine learning (ML) is not without its difficulties, particularly when the feature space becomes highly dimensional. While most ML models operate with the conventional Euclidean distance, the hidden hierarchical structure present among a set of phylogenetically related sequences is difficult to represent in Euclidean space without losing much information or requiring many dimensions. Since such hierarchical structure can be informative to analysis tasks such as clustering and classification, we propose two measures for generating a distance matrix from a set of sequences based on distance in the Poincare disk model of hyperbolic geometry, or the Poincare distance, for short. Such a distance measure can allow to embedding of even a fully resolved phylogenetic tree in just two dimensions with minimal distortion to any hierarchical structure. Our first approach is based purely on the classical Poincare distance, while the other approach modifies this distance by combining the Euclidean norms and the dot product between the sequence representations. A thorough analysis of both measures demonstrates its superiority in a variety of genomic and proteomic sequence classification tasks in terms of efficiency, accuracy, predictive performance, and the capacity to capture significant sequence correlations. These approaches perform better than existing state-of-the-art methods across the majority of evaluation metrics.

genomics↗

bioRxiv2024

Compression and k-mer based Approach For Anticancer Peptide Analysis

Our research delves into the imperative realm of anti-cancer peptide sequence analysis, an essential domain for biological researchers. Presently, neural network-based methodologies, while exhibiting precision, encounter challenges with a substantial parameter count and extensive data requirements. The recently proposed method to compute the pairwise distance between the sequences using the compression-based approach [26] focuses on compressing entire sequences, potentially overlooking intricate neighboring information for individual characters (i.e., amino acids in the case of protein and nucleotide in the case of nucleotide) within a sequence. The importance of neighboring information lies in its ability to provide context and enhance understanding at a finer level within the sequences being analyzed. Our study advocates an innovative paradigm, where we integrate classical compression algorithms, such as Gzip, with a pioneering k-mersbased strategy in an incremental fashion. Diverging from conventional techniques, our method entails compressing individual k-mers and incrementally constructing the compression for subsequences, ensuring more careful consideration of neighboring information for each character. Our proposed method improves classification performance without necessitating custom features or pre-trained models. Our approach unifies compression, Normalized Compression Distance, and k-mers-based techniques to generate embeddings, which are then used for classification. This synergy facilitates a nuanced understanding of cancer sequences, surpassing state-of-the-art methods in predictive accuracy on the Anti-Cancer Peptides dataset. Moreover, our methodology provides a practical and efficient alternative to computationally demanding Deep Neural Networks (DNNs), proving effective even in low-resource environments.

bioinformatics↗

bioRxiv2024

DeepPWM-BindingNet: Unleashing Binding Predictionwith Combined Sequence and PWM Features

A crucial challenge in molecular biology is the prediction of DNA-protein binding interactions, which has applications in the study of gene regulation and genome functionality. In this paper, we present a novel deep-learning framework to predict DNA-protein binding interactions with increased precision and interoperability. Our proposed framework DeepPWM-BindingNet leverages the rich information encoded in Position Weight Matrices (PWMs), which capture the sequence-specific binding preferences of proteins. These PWM-derived features are seamlessly integrated into a hybrid model of convolutional recurrent neural networks (CRNNs) that extracts hierarchical features from DNA sequences and protein structures. The sequential dependencies within the sequences are captured by recurrent layers. By incorporating PWM-derived features, the models interpretability is improved, enabling researchers to learn more about the underlying binding mechanisms. The models capacity to locate crucial binding sites is improved by the incorporation of an attention mechanism that highlights crucial regions. Experiments on diverse DNA-protein interaction datasets demonstrate the proposed approach improves the predictive performance. The proposed model holds significant potential in deciphering intricate DNA-protein interactions, ultimately advancing our comprehension of gene regulation mechanisms.

bioinformatics↗

bioRxiv2023

Hist2Vec: Kernel-Based Embeddings for Biological Sequence Classification

Biological sequence classification is vital in various fields, such as genomics and bioinformatics. The advancement and reduced cost of genomic sequencing have brought the attention of researchers for protein and nucleotide sequence classification. Traditional approaches face limitations in capturing the intricate relationships and hierarchical structures inherent in genomic sequences, while numerous machine-learning models have been proposed to tackle this challenge. In this work, we propose Hist2Vec, a novel kernel-based embedding generation approach for capturing sequence similarities. Hist2Vec combines the concept of histogram-based kernel matrices and Gaussian kernel functions. It constructs histogram-based representations using the unique k-mers present in the sequences. By leveraging the power of Gaussian kernels, Hist2Vec transforms these representations into high-dimensional feature spaces, preserving important sequence information. Hist2Vec aims to address the limitations of existing methods by capturing sequence similarities in a high-dimensional feature space while providing a robust and efficient framework for classification. We employ kernel Principal Component Analysis (PCA) using standard machine-learning algorithms to generate embedding for efficient classification. Experimental evaluations on protein and nucleotide datasets demonstrate the efficacy of Hist2Vec in achieving high classification accuracy compared to state-of-the-art methods. It outperforms state-of-the-art methods by achieving > 76% and > 83% accuracies for DNA and Protein datasets, respectively. Hist2Vec provides a robust framework for biological sequence classification, enabling better classification and promising avenues for further analysis of biological data.

bioinformatics↗

bioRxiv2023

Advancing Protein-DNA Binding Site Prediction: Integrating Sequence Models and Machine Learning Classifiers

Predicting protein-DNA binding sites is a challenging computational problem in the field of bioinformatics. Identifying the specific residues where proteins bind to DNA is of paramount importance, as it enables the modeling of their interactions and facilitates downstream studies. Nevertheless, the development of accurate and efficient computational methods for this task remains a persistent challenge. Accurate prediction of protein-DNA binding sites has far-reaching implications for understanding molecular mechanisms, disease processes, drug discovery, and synthetic biology applications. It helps bridge the gap between genomics and functional biology, enabling researchers to uncover the intricacies of cellular processes and advance our knowledge of the biological world. The method used to predict DNA binding residues in this study is a potent combination of conventional bioinformatics tools, protein language models, and cutting-edge machine learning and deep learning classifiers. On a dataset of protein-DNA binding sites, our model is meticulously trained, and it is then rigorously examined using several experiments. As indicated by higher predictive behavior with AUC values on two benchmark datasets, the results show superior performance when compared to existing models. The suggested model has a strong capacity for generalization and shows specificity for DNA-binding sites. We further demonstrated the adaptability of our model as a universal framework for binding site prediction by training it on a variety of protein-ligand binding site datasets. In conclusion, our innovative approach for predicting protein-DNA binding residues holds great promise in advancing our understanding of molecular interactions, thus paving the way for several groundbreaking applications in the field of molecular biology and genetics. Our approach demonstrated efficacy and versatility underscore its potential for driving transformative discoveries in biomolecular research.

bioinformatics↗

bioRxiv2023

Enhancing t-SNE Performance for Biological Sequencing Data through Kernel Selection

The genetic code for many different proteins can be found in biological sequencing data, which offers vital insight into the genetic evolution of viruses. While machine learning approaches are becoming increasingly popular for many "Big Data" situations, they have made little progress in comprehending the nature of such data. One such area is the t-distributed Stochastic Neighbour Embedding (t-SNE), a generalpurpose approach used to represent high dimensional data in low dimensional (LD) space while preserving similarity between data points. Traditionally, the Gaussian kernel is used with t-SNE. However, since the Gaussian kernel is not data-dependent, it determines each local bandwidth based on one local point only. This makes it computationally expensive, hence limited in scalability. Moreover, it can misrepresent some structures in the data. An alternative is to use the isolation kernel, which is a data-dependent method. However, it has a single parameter to tune in computing the kernel. Although the isolation kernel yields better performance in terms of scalability and preserving the similarity in LD space, it may still not perform optimally in some cases. This paper presents a perspective on improving the performance of t-SNE and argues that kernel selection could impact this performance. We use 9 different kernels to evaluate their impact on the performance of t-SNE, using SARS-CoV-2 "spike" protein sequences. With three different embedding methods, we show that the cosine similarity kernel gives the best results and enhances the performance of t-SNE.

bioinformatics↗