Research Theme 2
Exploring big data for new therapeutic strategy by computational methods

Genome-scale metabolic model-driven strategies to study gene expression data
 

Capturing disease mechanisms and identifying potential therapeutic targets through network analysis and machine learning algorithms 

Research on new cancer drugs is performed either through gene knockout studies or phenotypic screening of drugs in cancer cell lines. Both of these approaches are costly and time-consuming. A computational framework, e.g., genome-scale metabolic models (GSMMs), could be a good alternative to finding potential drug targets.

We first performed single-gene knockout studies on existing GSMMs of the NCI-60 cell lines obtained from 9 tissue types. The metabolic genes responsible for the growth of cancerous cells were identified and then ranked based on their cellular growth reduction. Gene ranking was used to identify potential drug targets that reduce the growth rate of cancer cells but not normal cells. The gene ranking results were also compared with existing shRNA screening data. We validated our theoretical results experimentally and showed that the drugs mitotane and myxothiazol can inhibit the growth of at least four cell‐lines of NCI‐60 database (Scientific Reports, 11:213 (2021)).

We also used GSMM to study the changes in metabolic fluxes related to metastatic transitions in ovarian cancer. We performed quantitative proteomics to identify protein signatures associated with three distinct phenotypic morphologies (2D monolayers and two geometrically distinct three-dimensional spheroidal states) of the high-grade serous ovarian cancer line OVCAR-3. We obtained disease-driving phenotype-specific metabolic reaction modules and elucidated gene knockout strategies to reduce metabolic alterations that could drive phenotypic transitions. Exploring the DrugBank database, we identified drugs that could impair such transitions and, hence, cancer progression. Finally, we experimentally validated our predictions by confirming the ability of one of our predicted drugs, the neuraminidase inhibitor oseltamivir, to inhibit spheroidogenesis in three ovarian cancer cell lines without any cytotoxic effects on untransformed stromal mesothelia (iScience, 26, 108081, (2023)).

GSMM was also used to delineate host response to Mycobacterium tuberculosis (Mtb) infection. The macrophage-like THP1 cells were infected with H37Ra, H37Rv, BND433, and JAL2287 Mtb strains, and host response was studied at 6, 18, 30, and 42 hours after infection. Integrating the temporal proteomics data in the genome-scale metabolic model (GSMM) gives context-specific GSMMs. We have developed a modified flux balance analysis (FBA), which doesn't require objective function to find the fluxes of metabolic reactions. We have also established a method of rewiring using GSMMs to explore potential strategies to change the flux state of virulent infected macrophages against their avirulent counterparts. Our methodology gives a correlation between different flux states, the extent of which was interpreted as the extent of rewiring (Molecular Omics, 17, 296-306, (2021)).

Metabolic characteristics of pancreatic β-cells in patients with type-2 diabetes (T2D) were also studied through GSMM. Analysis of these flux states shows a reduction in the mitochondrial fatty acid oxidation and mitochondrial oxidative phosphorylation pathways, leading to decreased insulin secretion in diabetes. We observed elevated reactive oxygen species (ROS) generation through peroxisomal fatty acid β-oxidation. In addition, cellular antioxidant defence systems were found to be attenuated in diabetes. Our analysis also uncovered the possible changes in the plasma metabolites in diabetes due to the β-cells failure. These efforts subsequently identified seven metabolites associated with cardiovascular disease (CVD) pathogenesis, thus establishing its link as a secondary complication of diabetes (Computers in Biology and Medicine, 144, 105365, (2022)).

Therapeutic strategies that target key molecules have yet to fulfill expected promises for most common malignancies. Major difficulties include incomplete understanding and validation of these targets in patients and single-pathway targeted approaches proving ineffective therapies for human malignancies. So, for target identification, it is important to understand the molecular cross-talk among key signaling pathways and how targeted agents may alter them. Such a departure from the traditional paradigm of studying single pathways to a more global approach will aid in designing novel therapeutics and also overcome the shortcomings of the existing therapeutic strategies.

The first study jointly explored three methods, namely protein-protein interaction (PPI) network analysis, metabolic networks (MN) analysis and machine learning (ML) algorithm to identify more plausible disease targets. Individually, each method has its limitations. For instance, PPI network analysis does not capture the perturbation of metabolic reactions. The MN analysis identifies critical reactions but cramps to connect them with other crucial pathways. ML can identify different biomolecules’ predictive capabilities but fails to capture the underlying mechanism. With this motivation, we have developed a systematic approach that first identifies the crucial proteins in the PPI network using a guilt-by-association approach, investigates their classification capability through ML algorithms, and then captures their effect on MN. We applied it to non-alcoholic steatohepatitis (NASH). We obtained a set of proteins in close proximity to the disease-associated proteins that can classify the disease, control the disease PPI network, and transform the disease metabolic flux state towards health (Physica A, 624, 128955. (2023)). 

In the next study, we developed a multi-layer relatedness (MLR) approach to uncover novel autophagy-related proteins involved in diabetic retinopathy (DR) (the study was supported through THSTI intramural Project). We constructed a prior knowledge-based network and identified the topologically significant novel disease-related candidate autophagic proteins (CAPs). Then, we evaluated their significance in a gene co-expression and a differentially-expressed gene (DEG) network. Finally, we investigated the proximity of CAPs to the known disease-related proteins. We identified three crucial autophagy-related proteins, TP53, HSAP90AA1, and PIK3R1. They are strongly related to multiple detrimental characteristics of DR, such as pericyte loss, angiogenesis, apoptosis, and endothelial cell migration, and hence may be used to prevent or delay the progression and development of DR. We evaluated one of the identified targets, TP53, in a cell-based model and found that its inhibition resulted in reduced angiogenesis in high glucose condition required to control DR (Gene, 866, 147339 (2023)).

Integrating differential co-expression and the functional relationships, primarily focusing on the source nodes, will open novel insights about disease progression as the source proteins could trigger signaling cascades, mostly because they are transcription factors, cell surface receptors, or enzymes that respond instantly to a particular stimulus. A thorough contextual investigation of these nodes could lead to a helpful beginning point for identifying potential causal linkages and guiding subsequent scientific investigations to uncover mechanisms underlying observed associations. Application of the proposed methodology in glioblastoma identified a set of regulators that could classify the disease, limit cell growth, promote survivability, affect immune infiltrations, and are associated with disease hallmarks (Computational Biology and Chemistry,109, 108024 (2024)). 

Addressing the complexity of tumor heterogeneity and advancing precision medicine are key objectives in current breast cancer research. In this study, we aim to deepen understanding of the disease complexities at the molecular subtype level. RNA sequencing data for luminal A, luminal B, HER2-positive, and triple-negative breast cancer (TNBC) subtypes, with corresponding normal samples, were obtained from the Gene Expression Omnibus (GEO) database. We identified modules containing unique and significantly correlated genes with each subtype using weighted gene co-expression network analysis (WGCNA). The prognostic and diagnostic potential of the identified crucial proteins for each subtype was evaluated through survival and ROC curve analyses. The identified crucial proteins for different subtype module exhibited significant associations with adverse survival outcomes in breast cancer patients and also demonstrated remarkable diagnostic efficacy in differentiating tumor and normal samples (Journal of Proteins and Proteomics, (vol. 15), pp. 329-345 (2024)).