Glossary

This comprehensive glossary defines bioinformatics, technical, and Lobster AI-specific terms used throughout the documentation and platform.

A

Affinity Proteomics Protein analysis method using antibody-based assays for targeted protein detection and quantification. Examples include Olink panels and antibody arrays. Typically has lower missing values (<30%) compared to mass spectrometry.

Agent Specialized AI component in Lobster AI that handles specific analysis domains (e.g., Single-cell Expert, Proteomics Expert). Agents use natural language understanding to execute appropriate tools and workflows.

Agent Factory Function Python function that creates and configures an agent instance. Specified in the agent registry for dynamic agent loading. Example: lobster.agents.transcriptomics.transcriptomics_expert.

Agent Handoff Architectural pattern in Lobster AI where agents transfer tasks to other agents based on analysis requirements. Coordinated by the supervisor agent using handoff tools. Example: supervisor hands off single-cell analysis to transcriptomics_expert.

Agent Registry (ComponentRegistry) System in Lobster AI that discovers and manages all available agents via Python entry points. Located in lobster/core/registry.py. Single source of truth for agent discovery (v1.0.0+).

Analysis Step IR (Intermediate Representation) Structured representation of analysis operations used for provenance tracking and notebook export. Contains operation details, parameters, code templates, and execution context. Required for reproducibility.

AnnData Annotated data structure used in Python bioinformatics for storing high-dimensional biological data. Contains expression matrix (X), observations (obs), variables (var), and additional metadata.

Annotation Process of assigning biological meaning to data elements, such as identifying cell types in single-cell data or functional categories for genes.

---

B

Batch Effect Technical variation in data caused by non-biological factors such as processing date, equipment, or experimental conditions. Must be corrected to avoid confounding biological signals.

Benjamini-Hochberg (BH) Multiple testing correction method that controls the False Discovery Rate (FDR). Used to adjust p-values when testing many hypotheses simultaneously.

Bulk RNA-seq RNA sequencing of entire tissue samples or cell populations, providing average expression across all cells. Contrasts with single-cell RNA-seq which measures individual cells.

---

C

Capability-Based Routing Architecture pattern in ContentAccessService (v0.2+) where providers register their capabilities and are selected based on query requirements. Enables flexible provider selection with priority-based fallback.

Cell Type Annotation Process of identifying and labeling cell populations in single-cell data based on marker gene expression patterns and biological knowledge. Can be automated (CellTypist) or manual (expert curation).

Clustering Computational method to group similar observations (cells, samples, genes) based on expression patterns. Common algorithms include Leiden, Louvain, and k-means.

Content Caching Two-tier caching system in Lobster AI (v0.2+) with session cache (in-memory, temporary) and workspace cache (filesystem, persistent). Provides 30-50x speedup for repeated access to publications and datasets.

ContentAccessService Unified service in Lobster AI (v0.2+) for accessing scientific content through five specialized providers: AbstractProvider, PubMedProvider, GEOProvider, PMCProvider, and WebpageProvider. Implements three-tier cascade (PMC XML → Webpage → PDF) with automatic fallback.

Coefficient of Variation (CV) Statistical measure of relative variability, calculated as standard deviation divided by mean. Used to assess technical reproducibility in proteomics.

ComBat Batch correction method that removes batch effects while preserving biological variation. Commonly used in transcriptomics analysis.

Count Matrix Two-dimensional data structure with features (genes/proteins) as rows and samples (cells/samples) as columns, containing quantified expression values.

---

D

DDA (Data-Dependent Acquisition) Mass spectrometry method where the instrument selects the most abundant ions for fragmentation and identification. Traditional approach for shotgun proteomics.

DIA (Data-Independent Acquisition) Mass spectrometry method that systematically fragments all ions in predefined windows. Often provides more comprehensive and reproducible protein identification.

DataManagerV2 Core data management system in Lobster AI that handles multiple modalities, provenance tracking, and analysis history. Supports various data backends and formats.

Differential Expression (DE) Statistical analysis to identify genes or proteins with significantly different expression levels between conditions or cell types.

Design Matrix Mathematical representation of experimental design used in statistical models. Encodes relationships between samples and experimental factors.

Docling Advanced PDF parsing library in Lobster AI (v0.2+) for extracting scientific content from publications. Achieves >90% Methods section detection rate, handles tables and formulas. Falls back to PyPDF2 (30% detection) when unavailable. Install: pip install lobster[docling].

Doublet Artifact in single-cell RNA-seq where two cells are captured and sequenced together, appearing as a single cell with unusually high gene counts.

Download Queue Orchestration system in Lobster AI for managing multi-agent downloads. Research agent discovers URLs and creates queue entries (PENDING), supervisor polls queue and hands off to data_expert, which downloads and updates status (IN_PROGRESS → COMPLETED/FAILED).

---

E

Empirical Bayes Statistical method that improves parameter estimation by borrowing information across features. Used in limma and other differential expression tools.

Enrichment Analysis Statistical method to identify biological pathways or processes that are over-represented in a gene or protein list.

---

F

False Discovery Rate (FDR) Expected proportion of false positives among rejected hypotheses. Commonly controlled at 5% (0.05) in genomics studies.

Feature Selection Process of selecting the most informative variables (genes/proteins) for analysis, such as highly variable genes in single-cell RNA-seq.

Fold Change Ratio of expression between two conditions. Often expressed as log2 fold change where 1 represents 2-fold increase and -1 represents 2-fold decrease.

---

G

GEO (Gene Expression Omnibus) NCBI database containing high-throughput genomics data. Lobster AI can directly download and analyze GEO datasets using GSE accession numbers.

GSEA (Gene Set Enrichment Analysis) Method for determining whether a defined set of genes shows statistically significant differences between biological states.

---

H

H5AD HDF5-based file format for storing AnnData objects. Standard format for single-cell data that preserves metadata and analysis results.

Handoff Process in Lobster AI where the supervisor agent transfers a query to a specialized agent based on the analysis type required.

Highly Variable Genes (HVG) Genes showing high variation across cells or samples, often selected for downstream analysis as they contain the most biological information.

---

I

Imputation Statistical method to estimate missing values in datasets. Particularly important in proteomics where 30-70% of values may be missing.

Integration Process of combining multiple datasets or omics layers to enable joint analysis. Includes batch correction and multi-omics integration.

---

L

Label-Free Quantification (LFQ) Mass spectrometry approach that quantifies proteins without chemical labeling, using peptide intensity measurements.

Leiden Clustering Community detection algorithm for clustering that often performs better than Louvain clustering for biological data.

Limma R package for analyzing gene expression data using linear models and empirical Bayes methods. Popular for differential expression analysis.

Log Transformation Mathematical transformation that converts multiplicative relationships to additive ones. Common preprocessing step in genomics (log2 or natural log).

---

M

MA Plot Scatter plot showing log fold change (M) vs average expression (A). Used to visualize differential expression results and identify bias.

Manual Curation Workflow Interactive workflow in Lobster AI for expert-guided cell type annotation, metadata harmonization, and quality assessment. Combines automated suggestions with domain expertise for improved biological accuracy.

Marker Genes Genes specifically or highly expressed in particular cell types or conditions. Used for cell type identification and validation.

MaxQuant Software platform for analyzing mass spectrometry proteomics data. Provides protein identification and quantification from raw MS data.

Missing at Random (MAR) Missing data mechanism where missingness depends on observed data but not on the missing values themselves.

Missing Not at Random (MNAR) Missing data mechanism where missingness depends on the unobserved values. Common in proteomics for low-abundance proteins.

Modality In Lobster AI, a named dataset representing a specific omics measurement (e.g., "single_cell_rna", "ms_proteomics"). Managed by DataManagerV2.

Moran's I Statistical measure of spatial autocorrelation used in spatial omics to identify spatially variable genes or proteins.

MuData Data structure for storing and analyzing multi-modal omics data. Extends AnnData to handle multiple measurement types simultaneously.

---

N

Normalization Process of adjusting data to remove technical variation and enable comparison between samples. Methods include library size normalization, quantile normalization, and TMM.

Notebook Export Feature in Lobster AI that exports complete analysis workflows as reproducible Jupyter notebooks. Uses Papermill for parameterization, includes provenance metadata, code snippets, and execution instructions. Access via /pipeline export command.

NPX (Normalized Protein eXpression) Log2-transformed and normalized protein abundance values used in Olink affinity proteomics assays. Already log2-scaled, so should not be log-transformed again during preprocessing.

---

O

Olink Commercial platform for targeted proteomics using proximity extension assays (PEA). Provides high-quality protein measurements with low missing values.

Ontology Structured vocabulary defining relationships between biological concepts. Gene Ontology (GO) is widely used for functional annotation.

---

P

PCA (Principal Component Analysis) Dimensionality reduction technique that identifies the directions of maximum variance in data. Used for visualization and quality control.

PDB (Protein Data Bank) Public repository of 3D structural data for proteins and nucleic acids. Lobster AI (v0.2+) fetches structures using 4-character PDB IDs (e.g., "1AKE") and links them to gene expression data. Supports both PDB and mmCIF formats.

PMC (PubMed Central) Free full-text archive of biomedical literature. Lobster AI (v0.2+) uses PMC XML API as priority source for publication content (500ms-2s response time, 30-40% coverage). Part of ContentAccessService three-tier cascade.

Protein Structure Visualization Feature in Lobster AI (v0.2+) for visualizing protein 3D structures using PyMOL. Supports interactive mode (GUI) and batch mode (PNG generation), residue highlighting, and automatic structure-to-gene linking. Install: make install-pymol.

Provider Component in ContentAccessService (v0.2+) that implements access to specific content sources. Five providers: AbstractProvider (abstracts, fast), PubMedProvider (literature search), GEOProvider (dataset discovery), PMCProvider (full-text, priority), WebpageProvider (fallback, PDF support).

Pseudobulk Method to aggregate single-cell data to sample-level summaries, enabling population-level statistical analysis with established bulk RNA-seq methods. Recommended for differential expression testing across conditions.

Publication Intelligence Automated system in Lobster AI (v0.2+) for extracting scientific methods and parameters from publications. Uses Docling for PDF parsing, ContentAccessService for content access, and structured schemas for metadata extraction.

Pseudotime Computational measure of cell progression along a biological process, such as differentiation or cell cycle, based on expression similarity.

pyDESeq2 Pure Python implementation of the DESeq2 algorithm for differential expression analysis of RNA-seq data.

PyMOL Professional molecular visualization system for protein structures. Integrated into Lobster AI (v0.2+) for 3D structure visualization. Supports both interactive GUI mode and headless batch mode (PNG generation). Install: make install-pymol or brew install brewsci/bio/pymol.

---

Q

Quality Control (QC) Assessment of data quality including technical metrics, batch effects, and sample integrity. Essential first step in any omics analysis.

Quantile Normalization Normalization method that makes the distribution of values identical across samples by matching quantiles.

---

R

RNA Velocity Method to predict future cell states by analyzing spliced and unspliced mRNA ratios, revealing cell differentiation dynamics.

---

S

S3 Backend Cloud storage integration in Lobster AI (v0.2+) for storing analysis data and results in AWS S3. Provides scalable storage for large datasets with seamless switching between local and cloud storage. Requires AWS credentials configuration.

scanpy Python package for analyzing single-cell gene expression data. Provides comprehensive tools for preprocessing, visualization, and analysis.

Service In Lobster AI architecture, stateless components that perform specific analysis tasks. Services receive AnnData objects and return results with statistics.

Session Export Feature in Lobster AI that exports complete analysis session including conversation history, loaded modalities, workspace content, generated plots, and provenance metadata. Access via /export session command. Enables session restoration and sharing.

Supervisor Agent Central coordinator agent in Lobster AI that routes user queries to specialized agents based on analysis requirements. Manages agent handoffs, monitors download queue, and orchestrates multi-agent workflows. Entry point for all user interactions.

Silhouette Score Measure of clustering quality that quantifies how similar objects are within clusters compared to other clusters.

Single-cell RNA-seq (scRNA-seq) Sequencing technology that measures gene expression in individual cells, revealing cellular heterogeneity and rare cell types.

Spatial Omics Analysis of molecular data with preserved spatial context, such as Visium spatial transcriptomics or imaging mass cytometry.

---

T

t-SNE Non-linear dimensionality reduction method that preserves local structure. Often used for visualizing high-dimensional single-cell data.

Three-Tier Cascade Content access strategy in ContentAccessService (v0.2+) that attempts multiple access methods with automatic fallback: (1) PMC XML API (priority, 500ms-2s, 30-40% coverage), (2) Webpage/PDF extraction (fallback, 2-8s, 60-70% coverage), (3) Error with alternative suggestions. Maximizes content accessibility.

TMM (Trimmed Mean of M-values) Normalization method commonly used in proteomics that assumes most proteins are not differentially expressed.

Trajectory Analysis Computational method to order cells along developmental or temporal progressions, revealing biological processes over time.

---

U

UMAP (Uniform Manifold Approximation and Projection) Dimensionality reduction and visualization technique that preserves both local and global data structure. Popular for single-cell visualization.

UMI (Unique Molecular Identifier) Short DNA sequence used to identify and count individual mRNA molecules, reducing PCR bias in single-cell sequencing.

---

V

Variance Stabilization Transformation that makes variance approximately constant across the range of data values. Important for statistical analysis assumptions.

Visium 10x Genomics platform for spatial gene expression analysis that measures transcriptomes with preserved spatial coordinates.

Volcano Plot Scatter plot showing fold change vs statistical significance (-log10 p-value) for differential expression results.

---

W

WGCNA (Weighted Gene Co-expression Network Analysis) Method for constructing gene co-expression networks and identifying modules of highly correlated genes.

---

Technical Terms (Lobster AI Specific)

CLI (Command Line Interface) Lobster's Rich-enhanced terminal interface with orange branding, autocomplete, and real-time monitoring capabilities.

Cloud Mode Operating mode where analyses are processed on cloud infrastructure. Activated by setting LOBSTER_CLOUD_KEY environment variable.

Handoff Tool Specialized tool registered in agent registry that allows supervisor agent to transfer tasks to specialist agents based on analysis requirements. Enables seamless multi-agent coordination with context preservation.

LangGraph Framework used by Lobster AI for creating multi-agent workflows with state management and tool integration.

Local Mode Default operating mode where all analyses are processed on the user's local machine.

Orange Theming Lobster AI's signature color scheme (#e45c47) used throughout the CLI interface and visualizations.

Provenance Tracking W3C-PROV compliant system in Lobster AI that records complete analysis history for reproducibility.

Rich CLI Enhanced command-line interface using the Rich Python library, providing advanced formatting, progress bars, and interactive elements.

Service Pattern Lobster AI's architectural pattern where stateless services handle analysis logic, returning both processed data and statistics.

Tool Function available to agents for performing specific tasks. Tools follow standardized patterns for data validation, service calls, and result storage.

Workspace Directory structure used by Lobster AI to organize data, results, and analysis history. Default location is .lobster_workspace/. Contains three subdirectories: literature/ (publications), data/ (datasets), metadata/ (custom content).

WorkspaceContentService Type-safe caching system in Lobster AI (v0.2+) for persistent storage of research content. Uses Pydantic schemas (PublicationContent, DatasetContent, MetadataContent) with enum-based validation (ContentType, RetrievalLevel). Provides identifier-based access to cached publications and datasets across sessions.

---

Statistical Terms

Adjusted Rand Score Measure of clustering performance that compares clustering results to known ground truth, corrected for chance.

Cohen's d Standardized measure of effect size representing the difference between groups in terms of standard deviations.

Empirical Bayes Moderation Statistical technique that improves variance estimates by borrowing information across genes, leading to more stable results.

Hypergeometric Test Statistical test used in enrichment analysis to determine if a pathway or gene set is over-represented in a list of interesting genes.

Multiple Testing Correction Statistical adjustment applied when testing multiple hypotheses simultaneously to control the overall error rate.

Pearson Correlation Measure of linear correlation between two variables, ranging from -1 (perfect negative correlation) to 1 (perfect positive correlation).

Spearman Correlation Rank-based correlation measure that captures monotonic relationships, more robust to outliers than Pearson correlation.

Wilcoxon Rank-Sum Test Non-parametric statistical test used to compare expression levels between groups. Default method for single-cell differential expression.

---

This glossary provides definitions for terms commonly encountered in Lobster AI documentation and bioinformatics analysis. For additional technical details, refer to the specific tutorial documents and API documentation.