What's New in Pipette: Smarter Analyses, Better Science

Pipette.bio Team

Feb. 28, 2026

We've been heads-down building for the past two months. Here's what changed — and why it matters for your research.

The Agent Now Shows Its Work

Every analysis Pipette runs involves dozens of decisions: which normalization to use, how to set thresholds, when to filter outliers. Previously, those decisions happened behind the scenes. Now, the agent's reasoning is visible in real time through an extended thinking scratchpad right in the UI.

You can watch as the agent evaluates your data, considers edge cases, and explains why it chose a particular approach. If something looks off, you'll see it as it happens — not after the pipeline finishes. This isn't just transparency for its own sake. When you can see the reasoning, you can catch mistakes earlier, learn from the agent's logic, and build more confidence in the results.

Every Analysis Gets Peer-Reviewed

This is probably the most important thing we shipped.

After your analysis completes, an independent methodology reviewer — a separate AI model — audits the entire pipeline. It checks for statistical errors, questionable parameter choices, missing controls, and biological plausibility. If it finds problems, it doesn't just flag them. The system automatically triggers remediation: re-running only the failed steps while preserving everything that passed.

This is DAG-aware, meaning if step 3 out of 12 had an issue, only step 3 and its downstream dependencies get re-executed. Your QC, alignment, and preprocessing stay intact. The reviewer's report is included with every result, so you always know exactly what was checked and what was fixed.

We think of it as a computational methods reviewer built into every analysis. The kind of scrutiny that usually happens months later during peer review now happens in minutes, before you even see the results.

Literature Context, Automatically

Before you even look at your results, Pipette now searches PubMed for papers relevant to your analysis — your organism, your genes of interest, your experimental design. It synthesizes this into a literature hypothesis: what's already known, what your results might confirm or challenge, and where the open questions are.

This isn't a generic literature review. It's grounded in your specific data and findings. If your differential expression analysis surfaces BRCA1 in a breast cancer dataset, the literature stage pulls the relevant clinical and mechanistic context automatically. It helps you interpret results faster and write up your findings with the right citations already in hand.

A Map of What Comes Next

Bioinformatics workflows are rarely one step. You run differential expression, then you want pathway enrichment. You do single-cell clustering, then you want trajectory analysis or cell-cell communication.

Pipette now understands these relationships through a skill graph — a directed network of 44 analysis types and 82 connections between them. After each analysis completes, the system suggests logical next steps based on what you just ran. Finished scRNA-seq clustering with Scanpy? You'll see suggestions for trajectory analysis, SCENIC regulatory networks, CellChat, and more — each with a one-line explanation of why it's relevant.

These are suggestions, not automatic executions. You stay in control of what runs. But you no longer need to know the full landscape of possible analyses to make good decisions about what to do next.

15 New Analysis Skills

We added support for a significant number of new analysis types:

• SCENIC — gene regulatory network inference and transcription factor activity scoring
• CellChat & CellPhoneDB — cell-cell communication from single-cell data
• RNA velocity — future cell state prediction from spliced/unspliced ratios
• Trajectory analysis — pseudotime ordering and lineage inference
• WGCNA — weighted gene co-expression network analysis for bulk RNA-seq
• Structural variants & CNV analysis — large-scale genomic alterations
• CRISPR screen analysis — hit calling from pooled CRISPR screens
• Bisulfite-seq — DNA methylation profiling
• Flux analysis — constraint-based metabolic modeling
• Clinical variant interpretation — with ClinVar and SnpEff integration
• Data inspection — lightweight file preview that runs on a smaller model to save you credits

Every skill follows the same pattern: download pre-built indexes from S3 when available, validate inputs before running, and produce a structured report with methods, results, and limitations.

Export Your Reports

Analysis reports can now be exported as PDF or DOCX directly from the UI. The exports include your methods section, results with figures, interpretation, and the methodology reviewer's audit — everything you need to drop into a manuscript supplement or share with collaborators. No copy-pasting from browser windows.

Dark Mode (Finally)

Pipette now has a proper light/dark theme toggle. The dark theme is the default — easier on the eyes during those late-night analysis sessions. Switch anytime from the sidebar.

What This Adds Up To

These aren't isolated features. They work together:

1. You upload data and start an analysis
2. The agent reasons through your problem with visible thinking
3. An independent reviewer audits the methodology and fixes issues automatically
4. Relevant literature is pulled and synthesized against your findings
5. The skill graph suggests what to explore next
6. You export the report as a PDF and move on

Every step that used to require manual effort or domain expertise is now handled — or at least suggested — by the system. The goal hasn't changed: make publication-quality bioinformatics accessible to every researcher, regardless of computational background.

Try it at pipette.bio. Your first 20 credits each month are free.