Single-Cell RNA Trajectory Inference: Creating a Multi-Panel Method Figure

Computational biology figures have a unique visual problem

Unlike wet-lab mechanism figures that show physical processes, computational biology figures must visualize abstract pipelines: data matrices, dimensionality reduction, clustering algorithms, and trajectory inference. The challenge is making these abstract algorithmic steps visually distinct and scientifically meaningful without falling into the common trap of generic flowchart boxes connected by plain arrows that communicate no biological or computational meaning.

The prompt

A bioinformatics researcher described their method:

The result

PaperFig generated a multi-panel ABCD layout — the standard format for computational method papers:

• Panel A: Data Acquisition — Single-cell isolation and RNA sequencing, showing the biological starting point.
• Panel B: Dimensionality Reduction & Embedding — High-dimensional gene expression matrix reduced to 2D embedding, with cell clusters visible as colored groups (T cells, B cells, myeloid cells, dendritic cells).
• Panel C: Trajectory Inference — Pseudotime trajectories connecting cell states, showing differentiation paths from root to terminal states with branching points.
• Panel D: Fate Interpretation — Marker gene expression patterns and immune cell fate assignments linked back to the trajectory.

Why the multi-panel ABCD format works for computational pipelines

Computational pipelines have a natural sequential structure that maps perfectly to the multi-panel format. The ABCD panel layout gives each algorithmic step its own dedicated visual space while maintaining a clear left-to-right or top-to-bottom reading order. Reviewers familiar with bioinformatics papers expect this specific format because it maps directly to how computational methods sections are written in the manuscript text.

For wet-lab biomedical mechanism figures rather than computational pipelines, the tumor organoid chip guide walks through a closely related multi-panel layout. For prompt-writing strategies that apply to any field, name your entities, their sequence, and their relationships explicitly before you describe visual style.

Tips for computational and bioinformatics figures

1. Name each pipeline stage. "Sequenced → embedded → trajectory inferred → marker genes → cell fate" gives PaperFig five distinct panels to populate.
2. Mention the data type. "Gene expression matrix" and "pseudotime" are specific enough for PaperFig to choose the right visual representation (heatmap, UMAP, trajectory plot).
3. Include the biological interpretation. "Immune cell fate" tells PaperFig to end with biology, not computation. This matches reviewer expectations that the method serves a biological question.
4. Request multi-panel layout. Computational papers almost always use ABCD panels. Mentioning "method figure" triggers this layout automatically.
5. Name the biological system. Stating "immune cell differentiation" or "tumor heterogeneity" rather than just "single-cell analysis" gives PaperFig the biological context to choose appropriate cell type labels and meaningful cluster annotations. Computational figures that lack biological grounding feel disconnected from the research question they are meant to address.

Prompt breakdown: why this description works

The prompt succeeds because it names every pipeline stage in temporal order. Each clause maps to one visual panel in the generated figure. Understanding this mapping helps you write better prompts for your own computational biology figures.

Data acquisition as the starting point

"Cells are sequenced" establishes the biological input. This grounds the figure in experimental reality rather than starting from a data matrix that appeared from nowhere. Reviewers want to see where the data came from, even in a computational methods figure.

Explicit algorithmic steps

"Gene expression matrix is embedded" and "pseudotime trajectory is inferred" name specific computational operations. These are not generic "processing" boxes — they are named methods with known visual representations (embedding plots, trajectory curves). The more specific you are about the algorithm, the more specific the visual output.

Biological interpretation as the endpoint

"Immune cell fate is interpreted" tells PaperFig to end with biology, not computation. This is a deliberate choice: computational biology papers must demonstrate that the pipeline serves a biological question. Ending with a UMAP plot and no interpretation is a common figure weakness that reviewers flag.

Adapting this prompt for other computational biology pipelines

The stage-by-stage structure works for any bioinformatics method:

Spatial transcriptomics pipeline: "Tissue sections are spatially barcoded, gene expression is mapped to coordinates, spatial domains are identified by clustering, ligand-receptor interactions are inferred between adjacent domains, and signaling networks are reconstructed."

Metagenomics pipeline: "Environmental samples are sequenced with shotgun metagenomics, reads are assembled into contigs, genes are predicted and annotated, metabolic pathways are reconstructed per genome bin, and community metabolic interactions are modeled."

Protein structure prediction: "Amino acid sequence is input, multiple sequence alignment identifies evolutionary conservation, attention-based transformer predicts structure, confidence scores are assigned per residue, and the predicted structure is validated against experimental data."

Each adaptation follows the same pattern: biological input, algorithmic stages in order, and biological or structural interpretation at the end.

Label conventions for computational biology figures

Data representation labels

• "UMAP" or "t-SNE" for dimensionality reduction plots (name the specific method, not "embedding")
• "Expression matrix" or "count matrix" for raw data
• "Cluster" with numbered or named identifiers (not just colored dots)
• "Pseudotime" with direction arrow showing the inferred temporal progression

Algorithm labels

• Name specific algorithms: "Monocle 3," "RNA velocity," "CellRank" rather than generic "trajectory inference"
• "Leiden clustering" or "Louvain clustering" rather than "community detection"
• "Highly variable genes" or "DEGs" rather than "important genes"

Biological labels

• Use standard cell type nomenclature from established atlases
• Distinguish "cell state" from "cell type" — they are different concepts
• Label branching points with the biological decision they represent (differentiation, activation, exhaustion)

Common mistakes in computational biology figures

Showing the code rather than the concept

Your figure should represent what the algorithm does, not how the code is structured. A box labeled "Python script" or "R function" tells the reader nothing about the scientific operation. Label with the biological or mathematical concept: "dimensionality reduction," "trajectory inference," "differential expression testing."

Using identical visual representations for different data types

A scatter plot can represent UMAP embeddings, gene expression correlations, or quality control metrics. If your figure has three scatter plots in a row with no visual distinction, the reader cannot tell what each represents. Use axis labels, color coding, and spatial context to differentiate panel types.

Omitting the validation step

Many computational pipeline figures show the method up to the prediction step and stop. Reviewers increasingly expect to see how the prediction is validated: comparison to ground truth, cross-validation, benchmarking against established methods. If validation is part of your paper, include it in the figure.

Forgetting the scale of the data

Computational biology datasets can range from hundreds to millions of cells or sequences. If dataset scale is relevant to your method's contribution (scalability), indicate it visually: sample size annotations, compute resource icons, or runtime comparisons.

No quality control steps

Every single-cell analysis pipeline includes quality control: filtering low- quality cells, removing doublets, normalizing for sequencing depth, correcting batch effects. If these steps are part of your contribution, include them in the figure. If they are standard preprocessing, consider showing them as a compact preprocessing block before the main analysis panels.

Ambiguous cell type assignments

If your figure shows colored clusters on a UMAP plot, every cluster should be labeled with a cell type or state. Unlabeled clusters suggest that you did not characterize them, which weakens the figure's scientific credibility. If a cluster is genuinely uncharacterized, label it as "unknown" or "unassigned" rather than leaving it unlabeled.

Missing the biological question

The most effective computational biology figures frame the analysis around a specific biological question: which cell type drives the disease? Where does the differentiation pathway branch? What signals correlate with treatment response? If your figure shows a pipeline without connecting it to a biological question, it reads as a technical demonstration rather than a scientific investigation.

FAQ

How do I show parallel processing in a pipeline figure?

Use a fan-out pattern: one input arrow splits into multiple parallel branches, each labeled with the specific analysis (methods extraction, data extraction, conclusion extraction). Show them converging again at the integration step. A multi-agent literature pipeline figure follows this exact pattern: name each agent, the analysis it performs, and the integration step where their outputs converge.

Should I include software tool logos in my figure?

Generally no. Tool logos add visual noise without scientific content. Name the tool in the panel label or figure caption instead. Exception: if your paper explicitly compares tools, showing their identifiers side by side may be appropriate.

What figure type works best for benchmarking papers?

Benchmarking figures typically use a split layout: Panel A shows the pipeline or method being evaluated, Panel B shows the benchmark results (bar charts, heatmaps, or radar plots). PaperFig handles the pipeline panel; use your data visualization tool for the benchmark results.

How do I represent uncertainty in trajectory inference?

Use dashed lines for uncertain trajectory edges, lighter colors for low- confidence assignments, or error bands around pseudotime estimates. Label the confidence metric explicitly so reviewers understand what the visual encoding represents.

Try it yourself

List your pipeline stages from raw data to biological interpretation, and PaperFig will structure them into a publication-ready method figure spanning sample processing, dimensionality reduction, trajectory inference, and marker validation.

Start creating for free →

Producing figures regularly? Compare plans →