Sequencing Results: RNA-Seq Analysis Tools

@
PROBLEM
Create an interface for scientists to receive their RNA sequencing results and reach experimental insights.
‍
SOLUTION
Built an interactive and performant suite of browser-based analysis tools. Users reduced their time-to-insights from weeks to minutes and removed the need for a dedicated bioinformatician to find top gene hits, confirm knockouts, and compare gene expression across different experimental conditions.
ROLE
Solo Designer & Researcher
Owned end-to-end design; partnered with PM and engineer as product team. Launched Oct 2025.
RESULTS
first 6 months
6x
growth in monthly orders
5x
growth in unique monthly customers
76%
customer return rate
Background
RNA sequencing has historically been used in limited scenarios (especially for smaller labs or researchers with less funding) because the time, expense, and expertise required for analysis is prohibitive. Despite limited use, it's a powerful assay because it shows how different treatments (e.g. adding a drug, or knocking out a gene) affect the transcription across the entire genome. Plasmidsaurus was able to take a process that previously cost $300–500 per sample and required 2–4 weeks of sequencing, plus 1–2 weeks for a dedicated bioinformatician to process and analyze, and turn it into a $50, 3-days-to-insights experience. This significantly opened up access to the power of RNA sequencing.

A significant value add for customers (aside from the 1000% increase in speed and 85% decrease in cost) was removing the need for a bioinformatician in order to interpret sequencing data. For our RNA-Seq product, I designed an intuitive analysis interface that allowed users to understand and evaluate their RNA sequencing data. My solution met the needs of users with varying RNAseq understanding and experience. Novices could complete their entire discovery analysis and walk away with a list of gene hits to follow up on, and seasoned experts could rapidly get an initial overview without needing to write any lines of code.
Design
There's many ways scientists can analyze their RNAseq data, which meant I needed to strike a balance between giving smart defaults and giving flexibility. I focused on an opinionated set of tools that would meet the analytical needs for 85% of our users, while introducing tweakable parameters and views to each of these tools. Our three main analysis tools allowed users to:

Interactive data exploration

RNAseq analysis is fundamentally a non-linear investigation. A researcher doesn't ask one question and move on. They ask a question, get a partial answer, and immediately need to interrogate that answer from a different angle. Unfortunately, RNAseq analysis results have historically been shared on static slides, which puts a huge obstacle to following up any additional investigative threads. I took advantage of our browser-based implementation and heavily used interactivity to connect the dots between our analysis tools. Users could follow up on threads of investigation in a single gesture.
The persistent gene picker at the top externalizes a user's working hypothesis: "These are the genes I am currently thinking about." The volcano plot on the left shows which genes are top hits for differential expression, and pathways with the most differential expression are bubbled to the top. Clicking on a functional pathway closes the loop for investigation by adding them into the gene picker.
By using a default of "30 Most Variable Genes" in the gene picker, users immediately land on interesting, explorable data when they open their results delivery. Saved gene lists allowed users to save investigative threads. The expression plot shown is optimized to be performant when pulling expression data from any of the ~20,000 genes (rows) across >100 samples (columns). 

AI Tools for Scientists

I conducted a research sprint to understand how users were already using LLMs with their data, what built and lost trust, and what ways agentic interactions could provide the most value. What emerged was that our users:
Based on these findings, I focused on designing an agentic chat that was an analytical co-pilot rather than an interpreter. The agent is able to compare with literature, explain analyses, even change the state of the page — filters, groupings, plot generation — but does not tell scientists what it means or what the next experimental step is. The focus was quickly expanding the surface of analysis (removing the manual step of putting together individual comparisons and plots) without generating scientific conclusions for the scientist.
Based on the user's instructions to follow up on a specific treatment condition (M1), the agent is able to identify the right set of comparisons, pull the top set of gene hits, and pull up the expression heatmap at the top of the page to show expression for relevant genes in that condition (M1) compared to a control (undifferntiated).
We also use LLMs to remove tedious tasks such as categorizing replicate samples into the same treatment category, or generating pairwise comparisons based on sample or category name.
Reflections
Users have responded very positively to the results page, citing the intuitiveness of interactions and unmatched speed-to-insight. Strong immediate adoption is one of the clearest validation signals a product can receive — it's clear that cheap, fast, analysis-ready RNA-Seq was a genuine gap in the market.

Future areas of iteration that I'm excited to expand on are:
  • Creating the concept of "Projects" to group together experimentally relevant RNA-Seq orders: this opens the door to larger time-course analyses, and allows users to build their narrative for writing a scientific paper or thesis directly through their sequencing orders.
  • Using experimental context provided in agentic chat to provide a more tailored experience for users: we can surface the most relevant plots, analyses, and QC considerations based on what questions we know the user is trying to answer.
A big learning from this project was how to work with moving constraints. While it's been incredible designing in parallel with active research from our bioinformatics and lab R&D team, it also meant what was technically feasible ended up being a bit of a moving goalpost. This required staying close to what our R&D groups were working on at any given moment, and building enough flexibility into design decisions to accommodate capability shifts without starting over. The tradeoffs can be challenging, but the result is so rewarding: the interface gives our users access to truly cutting-edge analyses.
Back to ProjectsLinkedInInstagram