Igniting full-length isoform analysis in single-cell and spatial RNA-seq data with FLAMESv2

FLAMESv2 is a highly modular, protocol-agnostic R/Bioconductor package designed to process and analyze long-read single-cell and spatial RNA-seq data, enabling comprehensive characterization of RNA isoforms, alternative splicing, and cellular heterogeneity across diverse experimental workflows.

The Gist

Imagine your body is a massive, bustling library. Inside this library, every cell is a unique reader holding a specific book. For a long time, scientists could only read the table of contents of these books. They could tell you which books were present (genes) and how many copies were on the shelf, but they couldn't see the actual stories inside.

This is because the old technology (short-read sequencing) was like a photocopier that could only snap a picture of the first page or the last page of a book. It missed the middle chapters, the plot twists, and the different endings.

However, genes are tricky. They often have multiple "editions" or isoforms. Think of a gene like a recipe for a cake. You can bake a vanilla cake, a chocolate cake, or a gluten-free version using the same basic ingredients. These different versions (isoforms) can do completely different jobs in the cell. Sometimes, a cell switches from baking a "growth" cake to a "repair" cake. If you only look at the table of contents, you miss these crucial changes.

Enter FLAMESv2: The Master Librarian

The paper introduces FLAMESv2, a new, super-smart software tool designed to read these entire books from cover to cover using "long-read" technology. Here is what makes it special, explained through some everyday analogies:

1. The Universal Translator (Protocol Agnostic)

Imagine you have a library with books written in different languages, some with weird binding styles, and some that are handwritten. Old tools were like translators who only spoke one specific language or only worked with hardcover books. If you gave them a paperback or a different language, they would crash.

FLAMESv2 is like a universal translator. It doesn't care if the data comes from a droplet-based machine, a spatial scanner, or a custom lab experiment. It can read the "structure" of the data and adapt. Whether you are looking at a single cell or a whole tissue map (spatial transcriptomics), FLAMESv2 can handle it.

2. The Modular Lego Kit

Many scientific tools are like a pre-built house: you can't move the walls or change the kitchen. If you want to do something slightly different, you have to tear the whole thing down.

FLAMESv2 is like a Lego set. It is built in "modules."

• Step 1: Sort the mail (Demultiplexing).
• Step 2: Read the address (Alignment).
• Step 3: Count the letters (Quantification).
• Step 4: Find new words (Isoform Discovery).

Because it's modular, scientists can swap out the "engine" for any step. If a new, better tool comes out for finding new words, they can plug it in without breaking the rest of the machine. It also lets you pause and resume work, so if your computer crashes, you don't have to start from scratch.

3. The Detective of Hidden Stories (Isoform Discovery)

In the past, scientists often had to guess what the middle of the story looked like. FLAMESv2 actually reads the whole thing.

The authors tested this on stem cells turning into neurons (like a caterpillar turning into a butterfly). They found that FLAMESv2 didn't just find the "standard" versions of the genes; it discovered thousands of new, unique story endings (novel isoforms) that previous tools missed.

They found that as cells matured, they didn't just turn genes "on" or "off." They switched versions of the genes. For example, they found a specific gene called PKM that had a "starter" version in young cells and a "mature" version in adult neurons. The mature version was missing a specific chapter (exon) that changed how the protein worked. This is the kind of detail that explains how a cell actually functions, not just what it looks like.

4. The Diversity Meter (Shannon's Entropy)

One of the coolest features is a new way to measure variety.
Imagine a cell is a chef.

• Low Diversity: The chef only makes one type of soup every day. (Low entropy).
• High Diversity: The chef makes a different soup every day, or mixes many ingredients in complex ways. (High entropy).

FLAMESv2 can measure this "soup variety" for every gene in every single cell. They found that some genes are very rigid (always the same soup), while others are incredibly flexible, with different cells making different "flavors" of the same gene. This helps scientists understand why two cells that look the same might actually behave very differently.

Why Does This Matter?

Before FLAMESv2, the world of single-cell long-read data was a mess of disconnected tools. You needed one program to sort the data, another to find the genes, and a third to visualize it, and they often didn't talk to each other.

FLAMESv2 brings order to the chaos. It is a single, fast, flexible toolkit that allows scientists to:

• See the full story of the cell's genetic instructions.
• Discover new biological "editions" of genes.
• Track how cells change over time (like development or disease).
• Do all this without needing a PhD in computer science to set it up.

In short, FLAMESv2 is the tool that finally lets us read the full, unedited manuscripts of life, revealing the hidden complexity that makes us who we are.

Technical Summary

1. Problem Statement

Long-read single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics have revolutionized the ability to profile RNA isoforms and alternative splicing at single-cell resolution. However, the analysis ecosystem for these data types remains highly fragmented and limited by several critical bottlenecks:

• Protocol Fragmentation: Existing tools are often tightly coupled to specific wet-lab protocols (e.g., specific 10x kits or custom barcoding methods), lacking flexibility for diverse read structures.
• Lack of End-to-End Pipelines: Many tools only address specific steps (e.g., demultiplexing or isoform identification) and require users to stitch together multiple software packages, often necessitating matched short-read data for barcode assignment or initial processing.
• Sparse Data Challenges: Isoform-level data is significantly sparser than gene-level data, requiring specialized, accurate quantification tools designed specifically for long-read data, which have only recently become available.
• Limited Visualization and Usability: Few tools offer comprehensive, publication-ready visualization or user-friendly interfaces for exploring isoform diversity and spatial distribution.

2. Methodology: FLAMESv2

The authors introduce FLAMESv2, a highly modular, protocol-agnostic R/Bioconductor package designed to provide an end-to-end pipeline for long-read single-cell and spatial RNA-seq analysis.

Core Pipeline Steps:

1. Barcode Demultiplexing: Utilizes the flexiplex and BLAZE packages to assign reads to cells (and spatial coordinates) based on user-defined read structures. It supports diverse protocols including 10x 3'/5', LR-Split-seq, Parse-like combinatorial barcoding, and spatial platforms (Visium, Curio).
2. Genome Alignment: Maps reads to a reference genome using minimap2.
3. Gene Quantification: Performs UMI deduplication to collapse duplicate molecules for accurate gene-level counting.
4. Isoform Identification: Generates an experiment-specific transcriptome reference. This step is tool-agnostic, allowing integration of discovery tools like bambu.
5. Transcriptome Alignment: Re-aligns deduplicated reads to the newly generated transcriptome reference.
6. Transcript Quantification: Quantifies isoform expression per cell. Supports probabilistic quantification of multi-mapping reads via oarfish or conservative counting via the original FLAMESv1 method.

Key Technical Features:

• Protocol Agnosticism: Users can define custom read structures via the flexiplex package, eliminating the need for hard-coded protocol support.
• Short-Read Independence: The pipeline can operate using long-read data alone, though it can also integrate matched short-read data if available.
• Modularity: Each step produces standard file formats (BAM, GTF, count matrices), allowing users to swap tools (e.g., changing the isoform quantifier) or skip steps (e.g., skipping novel isoform discovery to quantify only reference transcripts).
• Multi-Sample Support: Ensures consistent naming and annotation of novel isoforms across multiple samples in a single run, crucial for time-series or cohort studies.
• Performance: Optimized for speed, capable of processing >300 GB of compressed data (tens of millions of reads) within 24 hours.

3. Key Contributions

• Unified Framework: FLAMESv2 unifies the fragmented landscape of long-read scRNA-seq analysis into a single, flexible pipeline supporting single-cell, single-nuclei, spatial, and bulk protocols.
• Novel Diversity Metric: Introduces the find_diversity function, which calculates normalized Shannon's entropy per gene per cell. This metric quantifies whether a cell's expression of a gene is dominated by a single isoform (low entropy) or distributed across multiple isoforms (high entropy), enabling the study of intra-cellular isoform heterogeneity.
• Enhanced Visualization: Provides built-in functions for generating publication-ready figures, including isoform heatmaps, UMAPs colored by genotype or isoform usage, and spatial feature plots.
• Benchmarking Capability: The modular design allows for rigorous benchmarking of different sequencing protocols and computational tools within a unified framework.

4. Results

Benchmarking and Performance:

• 10x scRNA-seq: Benchmarked against scNanoSeq, SiCeLoRe, and wf-single-cell using the LongBench dataset. FLAMESv2 demonstrated the highest cell-to-cell gene expression concordance with short-read Illumina data (Spearman correlation ~0.496) and superior clustering accuracy (Adjusted Rand Index).
• Combinatorial Barcoding: Successfully processed LR-Split-seq data without matched short-reads, recovering all expected cell clusters and demonstrating a streamlined end-to-end workflow for Parse-like protocols.
• Spatial Transcriptomics: Applied to 10x Visium and Curio datasets. FLAMESv2 recapitulated known spatial distributions of cell types and achieved gene-level quantification accuracy comparable to or better than tools optimized for specific datasets (e.g., SiCeLoRe, spl-IsoQuant). It also successfully reproduced spatially specific isoform switching (e.g., Plp1).

Biological Applications:

• Neuronal Differentiation: Applied to an 80-day time-course of iPSC differentiation into neurons.
  • Isoform Discovery: Identified significantly more unique genes (29,013 vs. 5,578) and isoforms (93,721 vs. 15,430) compared to FLAMESv1, with a higher fraction of reads assigned to cells and a lower false-positive rate for novel isoforms (95% Full-Splice Matches vs. 27% in v1).
  • Developmental Trajectories: Reconstructed developmental trajectories from stem cells to radial glia to mature neurons. Identified 557 genes and 216 isoforms whose expression correlated with pseudotime.
  • Isoform Switching: Detected a specific isoform switch in the PKM gene, where progenitor cells expressed the canonical transcript and mature neurons expressed a novel isoform lacking exon 9, suggesting functional divergence.
• Isoform Diversity Analysis: Using Shannon's entropy on radial glial cells, the study revealed that while some genes (e.g., VIM) are dominated by a single isoform, others (e.g., RPL19, MAP1B) exhibit high intra-cellular diversity or inter-cellular heterogeneity in isoform proportions.

5. Significance

FLAMESv2 represents a major advancement in the analysis of long-read single-cell and spatial transcriptomics. By providing a modular, protocol-agnostic, and high-performance pipeline, it lowers the barrier to entry for researchers aiming to characterize RNA isoform biology.

• Scientific Impact: It enables the discovery of cell-type-specific isoform usage, developmental isoform trajectories, and intra-cellular heterogeneity that are invisible to short-read sequencing.
• Reproducibility: The unified framework addresses the reproducibility crisis in the field by standardizing analysis workflows across diverse experimental designs.
• Future Directions: The tool is poised to become essential for building isoform-resolved cell atlases and identifying isoform-level drivers of development and disease. The authors note future plans to integrate ambient RNA removal and support for emerging high-definition spatial platforms.