Galaxy-dl

Galaxy-dl recommends Galaxy tools by learning higher-order dependencies in workflows recorded on the European Galaxy server to support construction of bioinformatics analysis pipelines.

Key Features:

• Deep learning model: Employs a gated recurrent unit (GRU) neural network, a variant of recurrent neural networks (RNNs), to model sequence dependencies.
• Workflow representation and analysis: Analyzes workflows represented as directed acyclic graphs to identify patterns and sequences of tools used together.
• Tool recommendation and ranking: Predicts and ranks relevant Galaxy tools to aid pipeline construction, prioritizing commonly used and high-quality tools.
• Galaxy API integration: Exposes the recommendation model via the Galaxy API for programmatic access.
• Hyperparameter optimization and performance: Optimizes neural network hyperparameters using Bayesian optimization and reports a mean top-1 accuracy of 98%.

Scientific Applications:

• Workflow construction in bioinformatics: Supports assembly of analysis pipelines within the Galaxy ecosystem by recommending appropriate tools and sequences.
• Tool selection and reproducible pipeline design: Identifies frequently co-occurring tools to inform reproducible workflow composition and tool-choice decisions.

Methodology:

Trains a GRU-based RNN on workflows from the European Galaxy server using directed acyclic graph representations of workflows; optimizes neural network hyperparameters with Bayesian optimization and evaluates recommendations with a reported mean top-1 accuracy of 98%.