Kumar S, P. | bioRxiv Science

bioRxiv2024

GenomicSign: A computational method to discover unique, specific, and amplifiable signatures of target genomic sequences

Molecular diagnostics for the rapid identification of infectious, virulent, and pathogenic organisms are key to health and global security. Such methods rely on the identification and detection of signatures possessed by the organism. In this work, we outline a computational algorithm, GenomicSign, to determine unique and amplifiable genomic signatures of a set of target sequences against a background set of non-target sequences. The set of target sequences might comprise variants of a pathogen of interest, say SARS-CoV2 virus. Unique k-mers of the consensus target sequence for a range of k-values are determined, and the threshold k-value yielding a sharp transition in the number of unique k-mers is identified as kopt. Corresponding unique k-mers for k [≥] kopt are compared against the set of non-target sequences to identify target-specific unique k-mers. A pair of proximal such k-mers could enclose a potential amplicon. Primers to such pairs are designed and scored using a custom scheme to rank the potential amplicons. The top-ranked resulting amplicons are candidates for unique and amplifiable genomic signatures. The entire workflow is demonstrated using a case study with the SARS-CoV2 omicron genome. A case study distinguishing the SARS-CoV2 omicron target strain against non-target other SARS-CoV2 variants is performed to illustrate the workflow. GenomicSign has been implemented in Python and is available as an open-source software under MIT licence (https://www.github.com/apalania/GenomicSign).

bioinformatics↗

bioRxiv2023

Localgini: A method for harnessing inequality in geneexpression to improve the quality of context-specific models

Genome-scale metabolic models (GEMs) are valuable tools for investigating normal and disease phenotypes of biological systems through the prediction of fluxes in biochemical reactions. However, in specific contexts such as different cell lines, tissues, or diseases, only a subset of reactions is active. To address this, several model extraction methods (MeMs) have been developed to filter the reactions in GEMs and extract context-specific models. These methods utilize gene expression data as a source of context-specific information. To construct context-specific models, MeMs require core reactions specific to the given context as input. Typically, core reactions are derived using a single threshold applied to gene expression data. Reactions associated with genes whose expression values exceed the threshold are considered as core reactions. However, it is important to note that enzyme activity is not solely determined by gene expression levels. This approach based on a single threshold may inadvertently exclude reactions that require enzymes in smaller quantities. In this study, we propose a novel thresholding algorithm called Localgini, which leverages the Gini coefficient and transcriptomics data to derive gene-specific thresholds. Localgini is implemented as a pre-processing step to obtain core reactions for MeMs. To demonstrate the effectiveness of Localgini, we construct context-specific models for NCI-60 cancer cell lines and human tissues using different MeMs. We compare the performance of Localgini with existing thresholding methods, namely LocalT2 and StanDep. The results show that the models derived using Localgini recover a greater number of housekeeping functionalities compared to the other thresholding methods. Moreover, the Localgini-derived core reactions exhibit increased self-consistency and display enhanced consensus among models built using different MeMs. By incorporating transcriptomic support, Localgini includes low-expression reactions in the core reaction list, enhancing the comprehensiveness of the resulting models. Codes used in this study, compatible with COBRA toolbox are available at https://github.com/NiravBhattLab/Localgini Author summaryGenome-scale models are becoming a desirable tool to understand the metabolism of a biological system and hence find applications in the fields of systems and synthetic biology. These models are often integrated with transcriptomics data to improve prediction accuracy. Algorithms developed to integrate transcriptomics data with genome-scale models require core reactions to be derived from omics data using a threshold. In this work, we propose a thresholding method that uses an inequality-based metric to derive thresholds. We implied the proposed method and other existing methods to datasets of cancer cell lines and human tissue. We showed that our method improves the inclusion of reactions required for basic cellular maintenance. Furthermore, we validated the built models for the reduction in variance owing to the model-extraction algorithms. Overall, the proposed method improves the quality of metabolic models by inferring inequality in the distribution of gene expression levels across samples/contexts.

systems biology↗