graph attention neural networks  (SoftMax Inc)

Journal: Bioinformatics

Article Title: GOAT: Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network for eosinophilic asthma subtype

doi: 10.1093/bioinformatics/btad582

Figure Lengend Snippet: Schematic overview of GOAT. (A) Patient phenotype is the result of gene interactions regulating the interplay of multi-omics biomolecules. Our goal is to discover a network of genes that explains patient phenotype from multi-omics data. GOAT consists of two stages: prioritization of biomarker candidates using network propagation and identification of biomarkers using attention in graph transformer. (B) In the 1st stage, we prioritized genes that are important in discriminating phenotypes from phenotype-associated omics, proteome, and metabolome. Significant proteins and metabolites are identified and mapped to a gene-gene interaction graph. The triangle denotes the significant protein/metabolite features and the bold circle denotes the genes related to the significant features. Then network propagation prioritizes biomarker candidates. (C) In the 2nd stage, the protein–protein interaction network is trimmed with the genes prioritized in the 1st stage to make a network of biomarker candidates. Then the gene-level quantity of gene is given as node features (boxes next to the nodes) to generate graph instances of each patient. Multi-head graph attention from graph transformer model is trained to learn the latent representation of genes and the representations are concatenated to generate a patient vector. Lastly, multi-layer perception conducts a graph classification task, yielding genes with high attention weights as network biomarkers.

Article Snippet: We propose a deep attention model named Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network (GOAT) for identifying molecular biomarkers for eosinophilic asthma subtypes with multi-omics data.

Techniques: Biomarker Discovery, Plasmid Preparation

Journal: Bioinformatics

Article Title: GOAT: Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network for eosinophilic asthma subtype

doi: 10.1093/bioinformatics/btad582

Figure Lengend Snippet: Performance comparison with existing methods. Each dot in the plots depicts test AUPRC/AUROC over 10-fold CV. Boxplot comparing GOAT (orange) and existing multi-omics biomarker discovery methods (grey). The center line denotes the median, the upper and lower box boundaries denote upper and lower quartiles, and the whiskers denote 1.5× interquartile range. Denoted statistical annotations are retrieved from t -test (** P < .01). AUPRC, area under the precision–recall curve; AUROC, area under the receiver operating characteristic curve; CV, cross-validation; LR, logistic regression.

Article Snippet: We propose a deep attention model named Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network (GOAT) for identifying molecular biomarkers for eosinophilic asthma subtypes with multi-omics data.

Techniques: Comparison, Biomarker Discovery

Journal: Bioinformatics

Article Title: GOAT: Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network for eosinophilic asthma subtype

doi: 10.1093/bioinformatics/btad582

Figure Lengend Snippet: Power of using multi-omics network for biomarker discovery. Each dot in the plots depicts test AUPRC/AUROC over 10-fold CV. (A) Boxplot comparing feature selection method: DEG, DEP, and multi-omics NP. (B) Boxplot comparing the performance using multi-omics features (orange) and using single-omics features (green) in GNN model. When using single-omics features, only features from the specified omics were fed into the model. The center line denotes the median, the upper and lower box boundaries denote upper and lower quartiles, and the whiskers denote 1.5× interquartile range. Denoted statistical annotations are retrieved from t -test (* P < .05, ** P < .01, *** P < .001, **** P < .0001). AUPRC, area under the precision–recall curve; AUROC, area under the receiver operating characteristic curve; CV, cross-validation; DEG, differentially expressed gene; DEP, differentially expressed protein; multi-omics NP, multi-omics network propagation; GNN, graph neural network.

Article Snippet: We propose a deep attention model named Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network (GOAT) for identifying molecular biomarkers for eosinophilic asthma subtypes with multi-omics data.

Techniques: Biomarker Discovery, Selection