lobster-research 1.0.0

Research agents for Lobster AI - literature discovery and data management

lobster-research

Literature discovery and data acquisition agents for scientific research workflows.

Installation

pip install lobster-research

Agents

Agent	Description
research_agent	Literature discovery specialist. PubMed/bioRxiv search, GEO/SRA dataset discovery, metadata extraction, publication queue management.
data_expert_agent	Data operations specialist. Queue-based downloads, modality management, local file loading, workspace orchestration.

Services

Service	Purpose
ModalityDetectionService	Auto-detect data modality type from file characteristics

Features

Research Agent (Online Operations)

• PubMed literature search with filters and related paper discovery
• bioRxiv and medRxiv preprint search with full-text access
• GEO dataset discovery with organism and platform filtering
• SRA run metadata extraction and download URL generation
• PRIDE proteomics repository integration
• Full-text content extraction from PMC articles
• Methods section parsing for computational parameter discovery
• Publication queue for batch processing of research papers
• Automatic extraction of associated dataset identifiers

Data Expert Agent (Offline Operations)

• Execute downloads from pre-validated queue entries
• Zero online access boundary for security and reproducibility
• Multi-format file loading (CSV, TSV, H5AD, Excel)
• Modality listing, inspection, and validation
• Download strategy selection (AUTO, H5_FIRST, MATRIX_FIRST)
• Sample concatenation with union or intersection logic
• Failed download retry with exponential backoff
• Custom Python code execution for edge cases

Platform Support

• 10x Genomics MTX format (matrix, barcodes, features)
• H5AD pre-processed AnnData files
• Kallisto and Salmon bulk RNA-seq quantification
• CSV and TSV generic delimited matrices
• MaxQuant, Olink, and generic proteomics formats

Architecture

The research and data_expert agents implement a clean boundary pattern:

research_agent (ONLINE)              data_expert (OFFLINE)
-- Search literature                 -- Execute downloads
-- Discover datasets                 -- Load local files
-- Extract metadata/URLs             -- Manage modalities
-- Validate metadata                 -- Retry failed downloads
-- Create queue entries              -- Concatenate samples
         |                                    |
         ----------- Queue Entry -------------
                   (PENDING -> IN_PROGRESS -> COMPLETED)

Requirements

• Python 3.12+
• lobster-ai >= 1.0.0

Documentation

Full documentation: docs.omics-os.com/docs/agents/research

License

MIT