Single-cell RNA sequencing (scRNA-seq) is a powerful technique that allows scientists to study how genes are expressed in individual cells. This approach helps us understand how cells behave under different conditions—whether they’re fighting infections, responding to treatments, or affected by genetic changes. However, analyzing scRNA-seq data isn’t straightforward because there can be a lot of variability between cells, and it’s hard to tell which differences are biologically meaningful versus those caused by technical noise.

Memento, a new tool developed by researchers at UCSF, aims to improve the accuracy and efficiency of analyzing scRNA-seq data, helping researchers detect meaningful patterns in gene expression, variability, and interactions between genes—especially when working with massive datasets.

Why Is This Important?

When researchers compare cells from different experiments or conditions (like drug treatments or genetic changes), they need to understand which genes are genuinely affected and how the cells are responding. This means looking at several things, such as:

• Mean expression: Is the average level of a gene’s activity higher or lower in one group of cells compared to another?
• Variability: Are the levels of gene activity more or less stable across the group?
• Gene correlations: Are certain genes working together more (or less) under different conditions?

Memento makes it easier to answer these questions while reducing false positives (incorrect conclusions) and capturing subtle but important differences that other tools might miss.

Memento workflow for differential mean, variability, and gene correlation testing

(A) Experimental workflow for single-cell RNA sequencing (scRNA-seq) samples RNA transcripts inside each cell during library preparation and sequencing. After scRNA-seq sampling, patterns of mean, variability, and correlation of gene expression in the observed transcript counts no longer resemble the actual distribution. (B) Memento models scRNA-seq as a hypergeometric sampling process, estimates expression distribution parameters (mean, residual variance, and correlation) using method-of-moments estimators, implements efficient bootstrapping for estimating confidence intervals (CIs), and tests for differences in expression parameters between two groups of cells. (C) Four applications of Memento to characterize the response of ∼70,000 human tracheal epithelial cells to extracellular cytokines, reconstruct gene-regulatory networks from ∼170,000 human CD4+ T cells perturbed by CRISPR-Cas9, map the genetic determinants of gene expression in 1.2 M peripheral blood mononuclear cells (PBMCs) from 162 systemic lupus erythematosus (SLE) patients and 99 healthy controls, and comparisons of arbitrary groups of cells within the CELLxGENE Discover data corpus.

How Memento Was Tested

The researchers behind Memento tested it on several large datasets to show its value:

1. 70,000 tracheal epithelial cells – They identified genes activated by interferon, a molecule involved in immune responses to infections.
2. 160,000 T cells perturbed with CRISPR-Cas9 – They used this data to map how genes interact, helping reconstruct networks that control cell behavior.
3. 1.2 million peripheral blood mononuclear cells (PBMCs) – They looked for genetic variations (QTLs) specific to certain cell types.
4. 50 million-cell CELLxGENE Discover dataset – Memento compared different groups of cells, allowing researchers to explore patterns across this vast resource.

What Makes Memento Stand Out?

• More Significant and Reliable Results: Memento found meaningful changes in gene expression that other tools might overlook.
• Beyond Averages: It didn’t just look at average gene expression but also explored variability and relationships between genes, offering deeper insights.
• Scalability: It works efficiently with massive datasets, handling millions of cells and thousands of samples.

What Does This Mean for Future Research?

Tools like Memento give scientists new ways to analyze complex datasets, enabling them to uncover biological patterns with greater precision. Whether it’s understanding how cells respond to infections, identifying the best drug targets, or exploring the genetic differences between individuals, Memento offers a powerful way to advance research.

By using Memento, scientists can save time, reduce errors, and gain new insights into how cells function—ultimately helping to develop better treatments and improve healthcare outcomes.

Availability – Memento is implemented in Python, is compatible with scanpy, and can be downloaded at https://github.com/yelabucsf/scrna-parameter-estimation