Browse
Submit Data
Databases
API
Help

Dataset Information

0 Views

0 Connections

0 Citations

0 Reanalyses

0 Downloads

Omics score: 0

A scalable derivative-free optimizer trained through learning to optimize approach to interpret drug response mechanism network

ABSTRACT: With the emergence of various drug-based treatment strategies, many drug response prediction models have been developed to understand their effects. However, in order to gain comprehensive understanding of a drug response, a prediction model should reflect the underlying biological mechanisms, but the current models suffer from interpretability and scalability problems. Machine learning-based prediction models base their predictions on inferred features, which usually are not well correlated with biological mechanisms, posing a challenge on interpretability of its response predictions. In this regard, using Boolean modeling schemes may allow interpretations on mechanisms that contribute to a particular response, but optimizing Boolean models is difficult because of their high dimensional search space and discontinuous loss function. Here, we developed a scalable derivative-free optimizer for weighted sum Boolean network through meta-reinforcement learning. By using graph network and coordinate-wise policy, our learned optimizer can optimize high dimensional Boolean networks containing over 100 parameters of arbitrary structure, showing higher sample efficiency compared with other meta-heuristic algorithms. The optimized Boolean networks successfully predict the drug responses congruent with public databases and in-house experimental data. Moreover, mechanistic analysis of optimized networks shows reliable interpretability of the predictions by meaningful suggestions of known basket trial drug response prediction markers.

ORGANISM(S): Homo sapiens

PROVIDER: GSE184731 | GEO | 2024/04/27

REPOSITORIES: GEO

ACCESS DATA

Json Xml

Similar Datasets

Machine learning for discovery: deciphering RNA splicing logic

Project description:Machine learning methods, particularly neural networks trained on large datasets, are transforming how scientists approach scientific discovery and experimental design. However, current state-of-the-art neural networks are limited by their uninterpretability: despite providing accurate predictions, they cannot describe how they arrived at their predictions. Here, using an ``interpretable-by-design'' approach, we present a neural network model that provides insights into RNA splicing, a fundamental process in the transfer of genomic information into functional biochemical products. Although we designed our model to emphasize interpretability, its predictive accuracy is on par with state-of-the-art models. To demonstrate the model's interpretability, we introduce a visualization that, for any given exon, allows us to trace and quantify the entire decision process from input sequence to output splicing prediction. Importantly, the model revealed novel components of the splicing logic, which we experimentally validated. This study highlights how interpretable machine learning can advance scientific discovery.

2022-10-01 | GSE200096 | GEO

Drug combination sci-Plex Data

Project description:Recent advances in multiplexed single-cell transcriptomics experiments are facilitating the high-throughput study of drug and genetic perturbations. However, an exhaustive exploration of the combinatorial perturbation space is experimentally unfeasible, so computational methods are needed to predict, interpret, and prioritize perturbations. Here, we present the compositional perturbation autoencoder (CPA), which combines the interpretability of linear models with the flexibility of deep-learning approaches for single-cell response modeling. CPA encodes and learns transcriptional drug responses across different cell type, dose, and drug combinations. The model produces easy-to-interpret embeddings for drugs and cell types, which enables drug similarity analysis and predictions for unseen dosage and drug combinations. We show that CPA accurately models single-cell perturbations across compounds, doses, species, and time. We further demonstrate that CPA predicts combinatorial genetic interactions of several types, implying that it captures features that distinguish different interaction programs. Finally, we demonstrate that CPA can generate in-silico 5,329 missing genetic combination perturbations ($97.6% of all possibilities) with diverse genetic interactions. We envision our model will facilitate efficient experimental design and hypothesis generation by enabling in-silico response prediction at the single-cell level, and thus accelerate therapeutic applications using single-cell technologies.

2022-06-25 | GSE206741 | GEO

Analyzing synergistic and non-synergistic interactions in signalling pathways using Boolean Nested Effect Models

Project description:Understanding the structure and interplay of cellular signalling pathways is one of the great challenges in molecular biology. Boolean Networks can infer signalling networks from observations of protein activation. In situations where it is difficult to assess protein activation directly, Nested Effect Models are an alternative. They derive the network structure indirectly from downstream effects of pathway perturbations. To date, Nested Effect Models cannot resolve signalling details like the formation of signalling complexes or the activation of proteins by multiple alternative input signals. Here we introduce Boolean Nested Effect Models (B-NEM). B-NEMs combine the use of downstream effects with the higher resolution of signalling pathway structures in Boolean Networks. We show that B-NEMs accurately reconstruct signal flows in simulated data. Using B-NEM we then resolve BCR signalling via PI3K and TAK1 kinases in BL2 lymphoma cell lines. 84 BL2 cell-line samples were hybridized to HGU133+2 Affymetrix GeneChips.

2015-12-10 | E-GEOD-68761 | biostudies-arrayexpress

Analyzing synergistic and non-synergistic interactions in signalling pathways using Boolean Nested Effect Models

2015-12-10 | GSE68761 | GEO

Interpreting biologically informed neural networks for enhanced proteomic biomarker discovery and pathway analysis

Project description:The incorporation of machine learning methods into proteomics workflows improves the identification of disease-relevant biomarkers and biological pathways. However, machine learning models, such as deep neural networks, typically suffer from lack of interpretability. Here, we present a deep learning approach to combine biological pathway analysis and biomarker identification to increase the interpretability of proteomics experiments. Our approach integrates a priori knowledge of the relationships between proteins and biological pathways and biological processes into sparse neural networks to create biologically informed neural networks. We employ these networks to differentiate between clinical subphenotypes of septic acute kidney injury and COVID-19, as well as acute respiratory distress syndrome of different aetiologies. To gain biological insight into the complex syndromes, we utilize feature attribution-methods to introspect the networks for the identification of proteins and pathways important for distinguishing between subtypes. The algorithms are implemented in a freely available open source Python-package (https://github.com/InfectionMedicineProteomics/BINN).

2023-09-05 | PXD044264 | Pride

AI-guided use of balanced PPARα/g dual agonist finetunes macrophage responses in inflammatory bowel disease

Project description:A computational platform, the Boolean network explorer (BoNE), has recently been developed to infuse AI-enhanced precision into drug discovery; it enables querying and navigating invariant Boolean Implication Networks of disease maps for prioritizing high-value targets. Here we used BoNE to query an Inflammatory Bowel Disease (IBD)-map and prioritize two nuclear receptors, PPARa/g. Balanced agonism of PPARa/g was predicted to impact macrophage processes, ameliorate colitis in network-prioritized animal models, ‘reset’ the gene expression network from disease to health, and achieve a favorable therapeutic index that tracked other FDA-approved targets. Predictions were validated using a balanced and potent PPARa/g-dual agonist (PAR5359) in two pre-clinical murine models, i.e., Citrobacter rodentium-induced infectious colitis and DSS-induced colitis. Mechanistically, we show that such balanced dual agonism promotes bacterial clearance more efficiently than individual agonists both in vivo and in vitro; PPARa is required and its agonism is sufficient to induce the pro-inflammatory cytokines and cellular ROS, which are essential for bacterial clearance and immunity, whereas PPARg-agonism blunts these responses, delays microbial clearance and induces the anti-inflammatory cytokine, IL10. Balanced agonism achieved controlled inflammation while protecting the gut barrier and ‘reversal’ of the transcriptomic network. Furthermore, dual agonism effectively reversed the defective bacterial clearance observed in PBMCs derived from IBD patients. These findings not only deliver a macrophage modulator for use as barrier-protective therapy in IBD, but also highlight the potential of BoNE to accelerate and enhance the precision of drug discovery.

2022-02-03 | GSE171057 | GEO

Gene expression data of canine osteosarcoma primary tumors (Affymetrix Canine 2.0 array)

Project description:The recently developed COXEN method (PMID: 17666531) has been used to successfully extrapolate gene signatures of drug sensitivity across different tumor histotypes. We wanted to explore the utility of COXEN to predict chemosensitivity in canine cancer, specifically if we could extrapolate gene signatures identified in human datasets over to canine osteosarcoma tumors. This dataset of canine osteosarcoma tumor samples has available clinical outcome data after patients had infected limbs amputated and were treated with doxorubicin and/or carboplatin. We performed microarray analysis on this panel of tumor samples for validating our COXEN prediction models for doxorubicin or carboplatin sensitivity. Chemotherapy naive primary tumors were collected at the time of amputation and archived at the Flint Animal Cancer Center. RNA was extracted from 33 frozen archived tumor samples, followed by microarray analysis. The gene expression data was RMA preprocessed, scaled and were used as a independent test set to evaluate developed prediction models of sensitivity to doxorubicin or carboplatin. Drug predictions were than compared to clinical outcome in these patients that received doxorubicin and/or carboplatin.

2015-12-18 | E-GEOD-76127 | biostudies-arrayexpress

Causal Modeling Using Network Ensemble Simulations Predicts Novel Lipid Metabolism Genes

Project description:We describe a simulation strategy for prediction of drug targets and functional analysis of genetically defined associations. This strategy utilizes Bayesian model averaging over a sample of parameterized networks to achieve robust predictions in light of uncertainty of the network structure. We illustrate the method with an analysis of liver gene expression, serum lipid profiles and body weight measured on 120 male mice from a mouse intercross population. An ensemble of 1024 networks, gave accurate predictions of animals that were not part of the training data and explained almost twice the variance compared to quantitative trait loci alone. Additional in silico experiments identified 38 transcripts that are predicted to impact serum lipid profiles and suggest that they play a role in controlling high density lipoprotein and free triglycerides plasma concentrations.

2010-03-10 | GSE15226 | GEO

Plant Boolean Implication Network

Project description:Numerous gene expression datasets from diverse plant tissue samples have been already deposited in the public domain. There have been several attempts to do large scale meta-analyses of all of these datasets. Most of these analyses summarize pairwise gene expression relationships using correlation, or identify differentially expressed genes in two conditions. We propose here a new large scale meta-analysis of all of the publicly available plant datasets to identify Boolean logical relationships between genes. Boolean logic is a branch of mathematics that deals with two possible values. In the context of gene expression datasets we use qualitative high and low expression values. A strong logical relationship between genes emerges if at least one of the quadrants is sparsely populated.

2018-08-29 | GSE119083 | GEO

Human Boolean Implication Network

Project description:Numerous gene expression datasets from diverse human tissue samples have been already deposited in the public domain. There have been several attempts to do large scale meta-analyses of all of these datasets. Most of these analyses summarize pairwise gene expression relationships using correlation, or identify differentially expressed genes in two conditions. We propose here a new large scale meta-analysis of all of the publicly available human datasets to identify Boolean logical relationships between genes. Boolean logic is a branch of mathematics that deals with two possible values. In the context of gene expression datasets we use qualitative high and low expression values. A strong logical relationship between genes emerges if at least one of the quadrants is sparsely populated.

2018-08-29 | GSE119087 | GEO

OmicsDI is part of the ELIXIR infrastructure

OmicsDI is an Elixir interoperability service. Learn more ›

OmicsDI Databases

PRIDE
PeptideAtlas
MassIVE
JPOST Repository
Physiome Model Repository

EGA
EVA
ENA
LINCS
PAXDB
Cell Collective

MetaboLights
Metabolomics Workbench
MetabolomeExpress
GNPS
BioModels
FAIRDOMHub

ArrayExpress
dbGaP
ExpressionAtlas
GEO
NODE

Information

Databases
Help
API
Contact us
Code on GitHub
Terms of use
Submit Data