CelLAB: A Modular Image Analysis Platform for Quantitative Cell Biology with Embedded AI Assistant

Abstract

CelLAB (Cellular Analysis Workbench) is an open-source desktop platform for quantitative cell biology image analysis. It follows the workbench paradigm of engineering software, providing a unified environment where specialized analysis pipelines share a common project system, interactive GUI, and reusable computational engines. This release includes: Cornealia — Automated collective order analysis of corneal endothelial cell images from specular microscopy. Computes the radial distribution function g(r), interaction potential V(r), residual force field F(r), and the collective order biomarker K_ds, a physics-based metric that captures long-range spatial correlations and predicts corneal endothelium health better than conventional measures (cell density, hexagonality, CV). Includes mechanical disequilibrium analysis (mean residual force, compression fraction, hot spot detection), a signal-processing quality gate for automated reliability assessment, and publication-ready figure generation. Supports CellChek D (Konan Medical) machine output with automatic image cropping and OCR metric extraction. Full single-image and parallel batch processing with CLI and GUI workflows. Signal Associa — Coupling and causality analysis between biological time series. Quantifies directional and undirected associations using multiple complementary methods — Convergent Cross Mapping (CCM), Granger causality, Pearson and Spearman correlation, mutual information, and transfer entropy — with surrogate-based significance testing (IAAFT, AAFT, and shuffle nulls), automatic stationarization, and dual parametric/empirical p-values. Supports two batch modes: pairwise analysis of all variable pairs (producing coupling matrices and interaction networks) and pre-paired analysis of consecutive column pairs (producing a combined pairs summary). Includes multi-core CPU parallelization across pairs, per-run provenance (a README capturing inputs, methods, and surrogate settings), and publication-ready figures (null distributions, coupling matrices, significance matrices, and coupling networks). Full single-run and batch processing with GUI workflows. CellIntella (Beta — Experimental) — A privacy-preserving AI assistant embedded directly into CelLAB, provided as an experimental beta feature. Powered by Llama 3.2 3B running locally on CPU via llama.cpp, it provides context-aware guidance through a lightweight hierarchical retrieval-augmented generation (RAG) architecture. All conversations are fully private — no data leaves the user's machine, no API keys or internet connection required. CellIntella answers questions about Cornealia and Signal Associa parameters, analysis methods, quality gate interpretation, batch processing, CLI usage, and the CelLAB platform itself. As an early-stage feature, its responses should be verified against the bundled documentation. Six packages are provided: CelLAB_Win64 — Cornealia + Signal Associa + EasyOCR + GUI (Windows 10+, 64-bit) CelLAB_AI_Win64 — Full platform with CellIntella AI, Beta/Experimental (Windows 10+, 64-bit) CelLAB_Linux64 — Cornealia + Signal Associa + EasyOCR + GUI (Linux 64-bit, Ubuntu 20.04+) CelLAB_AI_Linux64 — Full platform with CellIntella AI, Beta/Experimental (Linux 64-bit, Ubuntu 20.04+) CelLAB_MacOS_ARM64 — Cornealia + Signal Associa + EasyOCR + GUI (macOS Apple Silicon) CelLAB_AI_MacOS_ARM64 — Full platform with CellIntella AI, Beta/Experimental (macOS Apple Silicon) No installation required — extract and run. No GPU required; all computation runs on CPU. The application runs through your web browser, and Python and all dependencies are bundled inside the package — no separate Python installation needed. System Requirements: 4 GB RAM minimum (8 GB recommended for AI version) Windows 10+ (64-bit), Ubuntu 20.04+ (64-bit), or macOS 13+ (Apple Silicon) No internet connection required for analysis or AI assistant Source Code: https://github.com/HanLab-BME-MTU/CelLAB