SCassist: An AI Based Workflow Assistant for Single-Cell Analysis

Abstract

Single-cell RNA sequencing (scRNA-seq) data analysis often involves complex iterative workflow, requiring significant expertise and time. To navigate this complexity, we have developed SCassist, an R package that leverages the power of the large language models (LLM's) to guide and enhance scRNA-seq analysis. SCassist integrates LLM's into key workflow steps, to analyze user data and provide relevant recommendations for filtering, normalization and clustering parameters. It also provides LLM guided insightful interpretations of variable features and principal components, along with cell type annotations and enrichment analysis. SCassist provides intelligent assistance using popular LLM's like Google's Gemini, OpenAI's GPT and Meta's Llama3, making scRNA-seq analysis accessible to researchers at all levels.

Figure 1:

The general architecture of the SCassist algorithm. SCassist, an LLM-powered assistant, streamlines single-cell analysis within the standard Seurat workflow. The top portion of the figure depicts the typical Seurat steps (quality control, normalization, dimensionality reduction, clustering, and annotation), while the interconnected pink boxes represent SCassist components, providing data-driven insights and parameter recommendations for each step. SCassist could be used at any stage of the standard single-cell workflow, starting from the quality control stage, where the user input for SCassist is simply the Seurat object containing the raw count matrix data. For the given Seurat object, SCassist generates metrics like summary statistics, quantile data, variance explained, and others. These metrics are then used to build augmented prompts for large language models (LLMs), recommending optimal parameters for filtering, normalization, dimensionality reduction, identifying significant features and offering insights (from variable genes, principal components, differentially expressed genes), and annotating clusters along with detailed reasoning.