Abhirupa Ghosh
Abhirupa Ghosh has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Sudipto Saha
Bose Institute
4 shared publications - Janani Ravi
Department of Biomedical Informatics, Center for Health Artificial Intelligence, University of Colorado Anschutz , Aurora, CO 80045
3 shared publications - Charmie K. Vang
University of Colorado Anschutz
2 shared publications - Emily A Boyer
University of Colorado Anschutz
2 shared publications - Ethan P Wolfe
University of Colorado Anschutz
2 shared publications - Evan Pierce Brenner
Department of Biomedical Informatics, Center for Health Artificial Intelligence, University of Colorado Anschutz , Aurora, CO 80045
2 shared publications
Latest Publications
- Open Access
Systematic evaluation using SigMatch reveals how sample tissues shape transcriptomic disease relationships
Motivation Diseases often share underlying molecular characteristics captured by transcriptional responses. Leveraging this knowledge transfer can aid in uncovering novel disease biology and drug repurposing. Large-scale transcriptomic datasets and disease signatures provide an opportunity to computationally identify these shared disease relationships based on shared genome-wide gene expression changes. However, disease signatures are strongly influenced by sample tissue, making it difficult to distinguish shared disease biology from shared tissue origin. Despite this challenge, existing computational approaches do not systematically take into account or evaluate the impact of tissue on their predictions. Results We present SigMatch, a systematic evaluation of the impact of sample tissue on transcriptomic-based disease relationship prioritization. We evaluate this impact on existing correlation- and rank-based methods, and on an alternative family called reconstruction-based methods. In particular, we examine (i) mechanistic disease relationships given the context of sample tissues and (ii) prediction agreement with external evidence of disease relatedness based on shared genes, symptoms, drugs, and comorbidities. We found that predicted disease relationships by all the methods are heavily influenced by sample tissue. Tissue-specific genes are not the sole driver of the prediction scores. Across methods, removal of tissue-specific genes did not reduce the observed tissue-dependent patterns or improve recovery of evidence-supported disease pairs. This work demonstrates substantial sample-tissue influence in transcriptomic similarities between diseases and that sample tissue should be considered when identifying and interpreting inferred transcriptomic disease relationships.
Open AccessSystematic evaluation using SigMatch reveals how sample tissues shape transcriptomic disease relationships
Motivation Diseases often share underlying molecular characteristics captured by transcriptional responses. Leveraging this knowledge transfer can aid in uncovering novel disease biology and drug repurposing. Large-scale transcriptomic datasets and disease signatures provide an opportunity to computationally identify these shared disease relationships based on shared genome-wide gene expression changes. However, disease signatures are strongly influenced by sample tissue, making it difficult to distinguish shared disease biology from shared tissue origin. Despite this challenge, existing computational approaches do not systematically take into account or evaluate the impact of tissue on their predictions. Results We present SigMatch, a systematic evaluation of the impact of sample tissue on transcriptomic-based disease relationship prioritization. We evaluate this impact on existing correlation- and rank-based methods, and on an alternative family called reconstruction-based methods. In particular, we examine (i) mechanistic disease relationships given the context of sample tissues and (ii) prediction agreement with external evidence of disease relatedness based on shared genes, symptoms, drugs, and comorbidities. We found that predicted disease relationships by all the methods are heavily influenced by sample tissue. Tissue-specific genes are not the sole driver of the prediction scores. Across methods, removal of tissue-specific genes did not reduce the observed tissue-dependent patterns or improve recovery of evidence-supported disease pairs. This work demonstrates substantial sample-tissue influence in transcriptomic similarities between diseases and that sample tissue should be considered when identifying and interpreting inferred transcriptomic disease relationships.
- Open Access
amR: an R package suite to predict antimicrobial resistance in bacterial pathogens
Motivation: Identifying bacterial antimicrobial resistance (AMR) is critical for diagnostics and treatment, but resistance is a complex trait arising from myriad mechanisms spanning multiple molecular scales. Existing computational approaches often function as black boxes and rarely explore cross-species or multi-drug patterns. We developed amR, an integrated R package suite that provides a complete framework from bacterial genome data curation to interpretable AMR predictions, enabling identification of resistance mechanisms across species and drugs. Results: The amR R package suite contains three modular packages. amRdata downloads genomes and paired antimicrobial susceptibility testing data from BV-BRC and processes them, constructs pangenomes, and extracts features at gene/protein cluster, protein domain, annotated Clusters of Orthologous Groups and ResFinder AMR-associated features, and structural variant scales; data are stored in memory-efficient formats (Parquet, DuckDB). amRml trains interpretable machine learning models per species-drug combination, calculates feature importance and performance metrics, and provides rich ground for hypothesis generation and mechanism discovery. amRviz provides an interactive Shiny dashboard to explore metadata distributions and model performance across species and drugs, visualize top predictive AMR features, and analyze cross-model patterns across geographic/temporal strata. We apply the suite to Shigella sonnei, achieving a median Matthews Correlation Coefficient of 0.89 across 23 drugs and drug classes. With thousands of genomes, multi-scale features, and interpretable models, amR provides an accessible, comprehensive framework for AMR research. The amR package suite is installable via GitHub (https://github.com/JRaviLab/amR; BSD-3-Clause license).
Open AccessamR: an R package suite to predict antimicrobial resistance in bacterial pathogens
Motivation: Identifying bacterial antimicrobial resistance (AMR) is critical for diagnostics and treatment, but resistance is a complex trait arising from myriad mechanisms spanning multiple molecular scales. Existing computational approaches often function as black boxes and rarely explore cross-species or multi-drug patterns. We developed amR, an integrated R package suite that provides a complete framework from bacterial genome data curation to interpretable AMR predictions, enabling identification of resistance mechanisms across species and drugs. Results: The amR R package suite contains three modular packages. amRdata downloads genomes and paired antimicrobial susceptibility testing data from BV-BRC and processes them, constructs pangenomes, and extracts features at gene/protein cluster, protein domain, annotated Clusters of Orthologous Groups and ResFinder AMR-associated features, and structural variant scales; data are stored in memory-efficient formats (Parquet, DuckDB). amRml trains interpretable machine learning models per species-drug combination, calculates feature importance and performance metrics, and provides rich ground for hypothesis generation and mechanism discovery. amRviz provides an interactive Shiny dashboard to explore metadata distributions and model performance across species and drugs, visualize top predictive AMR features, and analyze cross-model patterns across geographic/temporal strata. We apply the suite to Shigella sonnei, achieving a median Matthews Correlation Coefficient of 0.89 across 23 drugs and drug classes. With thousands of genomes, multi-scale features, and interpretable models, amR provides an accessible, comprehensive framework for AMR research. The amR package suite is installable via GitHub (https://github.com/JRaviLab/amR; BSD-3-Clause license).
- Open Access
From sequence to signature: Machine learning uncovers multiscale feature landscapes that predict AMR across ESKAPE pathogens
Since the clinical introduction of antibiotics in the 1940s, antimicrobial resistance (AMR) has become an increasingly dire threat to global public health. Pathogens acquire AMR much faster than we discover new drugs (antibiotics), warranting innovative methods to better understand its molecular underpinnings. Traditional approaches for detecting AMR in novel bacterial strains are time-consuming and labor-intensive. However, advances in sequencing technology offer a plethora of bacterial genome data, and computational approaches like machine learning (ML) provide an optimistic scope for in silico AMR prediction. Here, we introduce a comprehensive multiscale ML approach to predict AMR phenotypes and identify AMR molecular features associated with a single drug or drug family, stratified by time and geographical locations. As a case study, we focus on a subset of the World Health Organization’s Bacterial Priority Pathogens, the frequently drug-resistant and nosocomial ESKAPE pathogens: Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species. We started with sequenced genomes with lab-derived AMR phenotypes, constructed pangenomes, clustered gene and protein sequences, and extracted protein domains to generate pangenomic features across molecular scales. To uncover the molecular mechanisms behind drug-/drug class-specific resistance, we trained logistic regression ML models on our datasets. These yielded ranked lists of AMR-associated genes, proteins, and domains. In addition to recapitulating known AMR features, our models identified novel candidates for experimental validation. The models were performant across molecular scales, data types, and drugs while achieving a median normalized Matthews correlation coefficient of 0.89. Prediction performance showed resilience even when evaluated on geographical and temporal holdouts. We also evaluated model generalizability and cross-resistance across the drug-/drug class-specific models cross-tested on other available drug-/drug class genomes. Finally, we uncovered multiple drug class resistance features using multiclass and multilabel models. Our holistic approach promises reliable prediction of existing and developing resistance in newly sequenced pathogen genomes, while pinpointing the mechanistic molecular contributors of AMR. All our models and results are available at our interactive web app, https://jravilab.org/amr .
Open AccessFrom sequence to signature: Machine learning uncovers multiscale feature landscapes that predict AMR across ESKAPE pathogens
Since the clinical introduction of antibiotics in the 1940s, antimicrobial resistance (AMR) has become an increasingly dire threat to global public health. Pathogens acquire AMR much faster than we discover new drugs (antibiotics), warranting innovative methods to better understand its molecular underpinnings. Traditional approaches for detecting AMR in novel bacterial strains are time-consuming and labor-intensive. However, advances in sequencing technology offer a plethora of bacterial genome data, and computational approaches like machine learning (ML) provide an optimistic scope for in silico AMR prediction. Here, we introduce a comprehensive multiscale ML approach to predict AMR phenotypes and identify AMR molecular features associated with a single drug or drug family, stratified by time and geographical locations. As a case study, we focus on a subset of the World Health Organization’s Bacterial Priority Pathogens, the frequently drug-resistant and nosocomial ESKAPE pathogens: Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species. We started with sequenced genomes with lab-derived AMR phenotypes, constructed pangenomes, clustered gene and protein sequences, and extracted protein domains to generate pangenomic features across molecular scales. To uncover the molecular mechanisms behind drug-/drug class-specific resistance, we trained logistic regression ML models on our datasets. These yielded ranked lists of AMR-associated genes, proteins, and domains. In addition to recapitulating known AMR features, our models identified novel candidates for experimental validation. The models were performant across molecular scales, data types, and drugs while achieving a median normalized Matthews correlation coefficient of 0.89. Prediction performance showed resilience even when evaluated on geographical and temporal holdouts. We also evaluated model generalizability and cross-resistance across the drug-/drug class-specific models cross-tested on other available drug-/drug class genomes. Finally, we uncovered multiple drug class resistance features using multiclass and multilabel models. Our holistic approach promises reliable prediction of existing and developing resistance in newly sequenced pathogen genomes, while pinpointing the mechanistic molecular contributors of AMR. All our models and results are available at our interactive web app, https://jravilab.org/amr .
- 1 April 2020
- Published by Elsevier in Computational Biology and Chemistry
- Vol. 85, 107208
Computational approach to target USP28 for regulating Myc
- 1 April 2020
- Published by Elsevier in Computational Biology and Chemistry
- Vol. 85, 107208
Computational approach to target USP28 for regulating Myc
- 1 January 2018
- Published by Bentham Science Publishers Ltd. in Current Topics in Medicinal Chemistry
- Vol. 18 (13) , 1123-1134
Modulating Interleukins and their Receptors Interactions with Small Chemicals Using In Silico Approach for Asthma
Asthma is a complex, heterogeneous, airway inflammatory disorder broadly classified into atopic (IgE mediated) and non-atopic asthma. Monoclonal antibodies (MAbs) and small chemical protein-protein interaction modulators (PPIMs) are targeted against interleukins (ILs), which play a critical role in asthma. Many MAbs are targeted against ILs and IgE. Anti IgE MAb (Omalizumab) and Anti IL-5 MAbs (Mepolizumab, Reslizumab) have only been approved by FDA. Most of the MAbs including Tracolizumab, Lebrikizumab, Anrukinzumab (Anti IL-13 MAb), and Brodalumab (Anti IL-17 MAb) are in different phases of clinical trials. Pascolizumab (Anti IL-4 MAb), however, has failed. These MAbs are expensive and may render adverse immune response. Thus, small chemical modulators targeting ILs and their receptors (IL-Rs) are being exploited computationally and further validated experimentally. The complex ILs and IL-Rs available in PDB are best suited for these types of studies. A large number of small chemical modulators against protein-protein interactions (PPIs) have been compiled in a few databases like TIMBAL, 2P2I DB and IPPIDB. Small chemical libraries are used for virtual screening to find novel modulators targeting IL-R binding interface on IL. Molecular dynamic simulations have been further used for disruption mechanism and kinetic studies. IL-2/IL-2R was targeted with clinically tested small molecule modulators like SP4206, and IL-2 levels were known to increase in non-atopic asthma. In the absence of experimentally known modulators against atopic asthma, computational tools are being explored. For example, IL-33 is a target for atopic asthma where IL-33 and its receptor complex structure is available in PDB. In summary, small chemical modulators against ILs are a complementary approach to MAbs and computational tools have been used for identifying these modulators for asthma.
- 1 January 2018
- Published by Bentham Science Publishers Ltd. in Current Topics in Medicinal Chemistry
- Vol. 18 (13) , 1123-1134
Modulating Interleukins and their Receptors Interactions with Small Chemicals Using In Silico Approach for Asthma
Asthma is a complex, heterogeneous, airway inflammatory disorder broadly classified into atopic (IgE mediated) and non-atopic asthma. Monoclonal antibodies (MAbs) and small chemical protein-protein interaction modulators (PPIMs) are targeted against interleukins (ILs), which play a critical role in asthma. Many MAbs are targeted against ILs and IgE. Anti IgE MAb (Omalizumab) and Anti IL-5 MAbs (Mepolizumab, Reslizumab) have only been approved by FDA. Most of the MAbs including Tracolizumab, Lebrikizumab, Anrukinzumab (Anti IL-13 MAb), and Brodalumab (Anti IL-17 MAb) are in different phases of clinical trials. Pascolizumab (Anti IL-4 MAb), however, has failed. These MAbs are expensive and may render adverse immune response. Thus, small chemical modulators targeting ILs and their receptors (IL-Rs) are being exploited computationally and further validated experimentally. The complex ILs and IL-Rs available in PDB are best suited for these types of studies. A large number of small chemical modulators against protein-protein interactions (PPIs) have been compiled in a few databases like TIMBAL, 2P2I DB and IPPIDB. Small chemical libraries are used for virtual screening to find novel modulators targeting IL-R binding interface on IL. Molecular dynamic simulations have been further used for disruption mechanism and kinetic studies. IL-2/IL-2R was targeted with clinically tested small molecule modulators like SP4206, and IL-2 levels were known to increase in non-atopic asthma. In the absence of experimentally known modulators against atopic asthma, computational tools are being explored. For example, IL-33 is a target for atopic asthma where IL-33 and its receptor complex structure is available in PDB. In summary, small chemical modulators against ILs are a complementary approach to MAbs and computational tools have been used for identifying these modulators for asthma.
- 1 December 2017
- Published by Elsevier in Biochimica et Biophysica Acta (BBA) - General Subjects
- Vol. 1861 (12) , 3190-3200
Calcimycin mediates mycobacterial killing by inducing intracellular calcium-regulated autophagy in a P2RX7 dependent manner
- 1 December 2017
- Published by Elsevier in Biochimica et Biophysica Acta (BBA) - General Subjects
- Vol. 1861 (12) , 3190-3200
Calcimycin mediates mycobacterial killing by inducing intracellular calcium-regulated autophagy in a P2RX7 dependent manner
- Open Access
- 19 April 2017
- Published by The Royal Society in Royal Society Open Science
- Vol. 4 (4) , 160501
PPIMpred: a web server for high-throughput screening of small molecules targeting protein–protein interaction
PPIMpred is a web server that allows high-throughput screening of small molecules for targeting specific protein–protein interactions, namely Mdm2/P53, Bcl2/Bak and c-Myc/Max. Three different kernels of support vector machine (SVM), namely, linear, polynomial and radial basis function (RBF), and two other machine learning techniques including Naive Bayes and Random Forest were used to train the models. A fivefold cross-validation technique was used to measure the performance of these classifiers. The RBF kernel of SVM outperformed and/or was comparable with all other methods with accuracy values of 83%, 79% and 90% for Mdm2/P53, Bcl2/Bak and c-Myc/Max, respectively. About 80% of the predicted SVM scores of training/testing datasets from Mdm2/P53 and Bcl2/Bak have significant IC50 values and docking scores. The proposed models achieved an accuracy of 66–90% with blind sets. The three mentioned (Mdm2/P53, Bcl2/Bak and c-Myc/Max) proposed models were screened in a large dataset of 265 242 small chemicals from National Cancer Institute open database. To further realize the robustness of this approach, hits with high and random SVM scores were used for molecular docking in AutoDock Vina wherein the molecules with high and random predicted SVM scores yielded moderately significant docking scores ( p -values < 0.1). In addition to the above-mentioned classification scheme, this web server also allows users to get the structural and chemical similarities with known chemical modulators or drug-like molecules based on Tanimoto coefficient similarity search algorithm. PPIMpred is freely available at http://bicresources.jcbose.ac.in/ssaha4/PPIMpred/ .
- 19 April 2017
- Published by The Royal Society in Royal Society Open Science
- Vol. 4 (4) , 160501
Open AccessPPIMpred: a web server for high-throughput screening of small molecules targeting protein–protein interaction
PPIMpred is a web server that allows high-throughput screening of small molecules for targeting specific protein–protein interactions, namely Mdm2/P53, Bcl2/Bak and c-Myc/Max. Three different kernels of support vector machine (SVM), namely, linear, polynomial and radial basis function (RBF), and two other machine learning techniques including Naive Bayes and Random Forest were used to train the models. A fivefold cross-validation technique was used to measure the performance of these classifiers. The RBF kernel of SVM outperformed and/or was comparable with all other methods with accuracy values of 83%, 79% and 90% for Mdm2/P53, Bcl2/Bak and c-Myc/Max, respectively. About 80% of the predicted SVM scores of training/testing datasets from Mdm2/P53 and Bcl2/Bak have significant IC50 values and docking scores. The proposed models achieved an accuracy of 66–90% with blind sets. The three mentioned (Mdm2/P53, Bcl2/Bak and c-Myc/Max) proposed models were screened in a large dataset of 265 242 small chemicals from National Cancer Institute open database. To further realize the robustness of this approach, hits with high and random SVM scores were used for molecular docking in AutoDock Vina wherein the molecules with high and random predicted SVM scores yielded moderately significant docking scores ( p -values < 0.1). In addition to the above-mentioned classification scheme, this web server also allows users to get the structural and chemical similarities with known chemical modulators or drug-like molecules based on Tanimoto coefficient similarity search algorithm. PPIMpred is freely available at http://bicresources.jcbose.ac.in/ssaha4/PPIMpred/ .
- Open Access
- 24 May 2016
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 11 (5) , e0155911
Computational Framework for Prediction of Peptide Sequences That May Mediate Multiple Protein Interactions in Cancer-Associated Hub Proteins
A considerable proportion of protein-protein interactions (PPIs) in the cell are estimated to be mediated by very short peptide segments that approximately conform to specific sequence patterns known as linear motifs (LMs), often present in the disordered regions in the eukaryotic proteins. These peptides have been found to interact with low affinity and are able bind to multiple interactors, thus playing an important role in the PPI networks involving date hubs. In this work, PPI data and de novo motif identification based method (MEME) were used to identify such peptides in three cancer-associated hub proteins—MYC, APC and MDM2. The peptides corresponding to the significant LMs identified for each hub protein were aligned, the overlapping regions across these peptides being termed as overlapping linear peptides (OLPs). These OLPs were thus predicted to be responsible for multiple PPIs of the corresponding hub proteins and a scoring system was developed to rank them. We predicted six OLPs in MYC and five OLPs in MDM2 that scored higher than OLP predictions from randomly generated protein sets. Two OLP sequences from the C-terminal of MYC were predicted to bind with FBXW7, component of an E3 ubiquitin-protein ligase complex involved in proteasomal degradation of MYC. Similarly, we identified peptides in the C-terminal of MDM2 interacting with FKBP3, which has a specific role in auto-ubiquitinylation of MDM2. The peptide sequences predicted in MYC and MDM2 look promising for designing orthosteric inhibitors against possible disease-associated PPIs. Since these OLPs can interact with other proteins as well, these inhibitors should be specific to the targeted interactor to prevent undesired side-effects. This computational framework has been designed to predict and rank the peptide regions that may mediate multiple PPIs and can be applied to other disease-associated date hub proteins for prediction of novel therapeutic targets of small molecule PPI modulators.
- 24 May 2016
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 11 (5) , e0155911
Open AccessComputational Framework for Prediction of Peptide Sequences That May Mediate Multiple Protein Interactions in Cancer-Associated Hub Proteins
A considerable proportion of protein-protein interactions (PPIs) in the cell are estimated to be mediated by very short peptide segments that approximately conform to specific sequence patterns known as linear motifs (LMs), often present in the disordered regions in the eukaryotic proteins. These peptides have been found to interact with low affinity and are able bind to multiple interactors, thus playing an important role in the PPI networks involving date hubs. In this work, PPI data and de novo motif identification based method (MEME) were used to identify such peptides in three cancer-associated hub proteins—MYC, APC and MDM2. The peptides corresponding to the significant LMs identified for each hub protein were aligned, the overlapping regions across these peptides being termed as overlapping linear peptides (OLPs). These OLPs were thus predicted to be responsible for multiple PPIs of the corresponding hub proteins and a scoring system was developed to rank them. We predicted six OLPs in MYC and five OLPs in MDM2 that scored higher than OLP predictions from randomly generated protein sets. Two OLP sequences from the C-terminal of MYC were predicted to bind with FBXW7, component of an E3 ubiquitin-protein ligase complex involved in proteasomal degradation of MYC. Similarly, we identified peptides in the C-terminal of MDM2 interacting with FKBP3, which has a specific role in auto-ubiquitinylation of MDM2. The peptide sequences predicted in MYC and MDM2 look promising for designing orthosteric inhibitors against possible disease-associated PPIs. Since these OLPs can interact with other proteins as well, these inhibitors should be specific to the targeted interactor to prevent undesired side-effects. This computational framework has been designed to predict and rank the peptide regions that may mediate multiple PPIs and can be applied to other disease-associated date hub proteins for prediction of novel therapeutic targets of small molecule PPI modulators.