CRISPR Screen Analysis
for Absolute Beginners
One recorded session that takes a real, published genome-wide screen from raw FASTQ files to a ranked list of genes and finished figures, on your own laptop. Every step explained, statistics included, and a pipeline that runs the whole thing on your own screen with two commands. No programming background assumed.
Six minutes on what the course is, who built it, and why “just ask AI” isn’t the answer it looks like.
Your screen came back.
Now what?
Months of work, seventy-seven thousand guides and a sequencing bill. Then the files come back and sit there, because nobody in the lab knows what to do with them.
The files sit in a folder nobody opens
Five FASTQ files, several gigabytes each, in a format nobody ever explained. The bioinformatician is booked for months, and the screen you spent half a year on waits behind everyone else’s.
Or a spreadsheet of gene names you can’t judge
The core sends back a ranked list. Is the top hit real, or one lucky guide? Did the screen even work? Without knowing how the list was made, you can’t tell — and you’re the one who’ll spend a month validating it.
You ask how to start, and get a reading list
Bash, Python, R, statistics, linear algebra — and only then your own data. I followed that list. It cost me about ten years, and most of it I have never used since.
Or you get the new answer: just ask AI
Upload the files, type “analyze my screen”, wait ten minutes. It’s certainly faster. The section below is what it cost when I tried it on the screen this course analyzes.
Ten minutes. No errors.
Wrong answer.
Ask an AI assistant how to run MAGeCK, the standard tool for these screens, and it will often tell you to use your non-targeting guides as controls. Plenty of tutorials say the same. I did exactly that, on the published screen this course analyzes.
- ✕Including 104 genes known not to be essential at all.
- ✕BAX, the paper’s main finding, no longer significant.
- ✕And not one error message. Everything ran, so nothing would have warned you.
That’s why I don’t believe the posts saying bioinformatics is dead. What’s dying is the pipeline-runner: the job of pushing files through a tool and passing on whatever comes out. AI does that now, mistakes included. What it can’t do is know your screen — which comparison answers your question, whether the essential genes dropped out, whether a hit rests on one lucky guide. That judgement is what this course teaches, and it’s the thing AI multiplies once you have it. Read the full argument →
I wasn’t a bioinformatician either.
I was a wet-lab researcher studying sex determination in turtles when our first sequencing data came back, and nobody in my department could analyze it, including me. I still remember sitting in front of files I couldn’t open, with nobody to ask.
I was handed the reading list, and I followed it. Years of it. Most of what was on it I have never used since, and the part I actually needed — what each step does to my data, and how to tell when a result is wrong — was never on the list at all.
Since then I’ve worked on hundreds of projects with more than two hundred collaborators, and screens are where I see the gap most often: expensive experiments, and nobody in the lab who can say whether the hit list is true. So I built this the way I wish someone had built it for me — starting from a laptop that has never run a command, and not skipping the parts that are supposed to be obvious.
So I teach it
the other way round.
No programming course, and no pipeline to build — it’s built for you. What you learn instead is what each step does and why, well enough to check the pipeline that does it for you.
Real data from the start
You analyze a real screen from the first step and understand it as you go, because feeling ready never arrives on its own. Aiming forever and never firing is how beginners lose years.
Exactly what a screen analysis needs
Not the whole of Linux, not a statistics degree. The handful of commands this analysis uses, and the statistics that decide what your results mean — learned at the moment you need them, and nothing more.
A data-fluent biologist
Fluent enough in your own data to direct the AI, check what it hands back, and catch a wrong answer before it costs you six months at the bench. That’s the goal, and it’s the skill that doesn’t get automated.
Every step of the analysis gets the same five answers. Nothing is shown without being explained.
One session. Three parts.
One complete analysis.
In order, nothing skipped. Each part assumes only the ones before it, and nothing at all is assumed before the opening. The example is a real published screen: a genome-wide Brunello knockout screen for resistance to the leukemia drug venetoclax (Nechiporuk et al., Cancer Discovery 2019, GEO GSE130413).
AI amplifies you, or it replaces you
The question everyone is asking — does AI make this skill pointless? — answered with real job-market numbers first, then with the test above, run on this very dataset. What you’ll skip on purpose, and four habits for using AI as a learning partner rather than a crutch: let it do the typing, and keep the decisions for yourself.
Linux on your own laptop, from the Microsoft Store
It doesn’t assume you’ve ever installed anything. On Windows, real Linux from the Microsoft Store, click by click, including what to do when the install fails. On a Mac, a better terminal and Homebrew. Then the Linux file system, the handful of commands you actually need, and how to read any command you meet. One script then installs and tests everything the analysis needs — MAGeCK, seqkit, and R with its packages — and you see what success looks like at each stage.
A real screen, from raw reads to biology
The biology first: how Cas9 knocks out a gene, how a pooled library of seventy-seven thousand guides gets into cells, and the idea the whole course rests on — every guide is a twenty-letter barcode, so a screen is a counting experiment. Then the data, one step at a time: inside a read, the library file, counting, the four numbers that tell you the screen worked, the normalization trap, how MAGeCK tests each guide and turns guides into genes, whether the essential genes dropped out, the drug comparison, and how candidates get picked. Every output opened in Excel.
The pipeline, pointed at your screen
Everything from Part 2 turned into a pipeline. You describe your experiment in three small text files — the settings, your samples, and the comparisons you want. The first command runs the whole MAGeCK workflow, checks your reads before counting and prints the quality table with flags. The second draws every figure in about half a minute, as PNGs for slides and PDFs for papers. Then a worked example for a typical drug screen with replicates, and a troubleshooting table for the errors people actually hit.
You don’t just learn the pipeline.
You leave with it.
The pipeline sits outside the session, because it’s the part you copy into your next project. The practice workbook sits beside it, for when you want to see each step work with your own eyes first.
CRISPR_Screen_Pipeline/
Three small settings files you edit, one command for the MAGeCK workflow and one for every figure. It takes your own library file in whatever layout it came in, handles mouse screens too, and runs on your laptop: counting the full example screen took about eight minutes. The course example’s five raw FASTQ files come with it, so it runs before you ever touch your own data.
├─ config/ · settings · samples · comparisons
├─ 03_run_mageck_workflow.sh → QC table
├─ 04_make_plots.R → every figure
└─ data/ · the example’s raw FASTQ files
The practice workbook
Optional, and there to help you understand rather than to analyze anything. Every step of Part 2 on a small practice dataset, each with a worked example, a “your turn” and the answer, and every command runs in seconds. The statistics exercises use the real count table, so the numbers match the video.
├─ every step of Part 2, in order
├─ worked example → your turn → answer
└─ its own small practice dataset
Start from raw FASTQ files or from a count table — the pipeline meets your data wherever it is.
One folder.
Everything in it.
So you always know where to look. Once you’re set up, all of it works offline, and all of it is yours to keep.
The video itselfOne session in three parts, 2 h 35 m, yours for as long as you want it
Every slide as a PDFAll 114, to keep, to search and to print
The commands handoutEvery command in the course, the file formats, every mageck test option, and links to every tool
The practice workbookWith its own small dataset and the answer after every exercise
The pipeline for your own screenWith the example’s raw FASTQ files bundled in — 6.5 GB, so nothing depends on a download working
A setup guide for Windows and MacAbout an hour, most of it unattended downloading, and you do it once
Every guide you double-click to openEach one as a page in your browser, and as plain text beside your code
Downloaded, not streamedEverything sits on your own disk, so nothing depends on a server staying up
Here’s how I’d actually do it.
Watch for understanding, then run it yourself. You don’t have to type along with the video.
Set up once — about an hour
Follow the setup guide for Windows or Mac, or set up alongside Part 1 of the video. You do it one time, and then it’s done.
Watch Part 2 for understanding
Every step is shown, explained and opened in front of you. You don’t need to type along; what you want to be watching is what each step is for, what goes in, what comes out, and how to tell when something has gone wrong.
Then the practice workbook, if you want hands-on
Every command runs in seconds on the small practice dataset, and the answer follows each exercise. It’s optional: it’s there so you can see each step work with your own eyes.
Run the pipeline on the course example, then on your own screen
Check your numbers match the video — a hundred and three dropouts, and BAX on top — and you know your setup is right. Then point it at your own data.
When something breaks — and it will
Copy the exact error with the command you ran, paste it into an AI, and ask what it means. That’s what I do. Missed a step, or want to look one up later? The same analysis is written up step by step in the free tutorial on NGS101 — use it as a dictionary. You should know before you buy: this course doesn’t come with me answering your questions. That’s the live workshop, and it’s a different thing.
See exactly what’s inside.
Ten minutes, the whole course part by part, what each one teaches and what you walk away with. Everything in it is a real slide out of the real deck — not a summary made for the video. If you’re still deciding, this should be enough to decide on.
Free, no email required. Watch it before you spend anything.
Who this is for —
and who it isn’t.
There are no prerequisites. If you have never opened a terminal, you’re not behind; you’re exactly on time.
This is for you if…
- ✓You’re a wet-lab scientist with a pooled CRISPR screen, finished or planned, and you want to analyze it and judge the result yourself.
- ✓You’ve never opened a terminal, or you’ve opened one a couple of times and nothing more.
- ✓A core or a collaborator sent you a ranked gene list, and you’d like to know whether to believe it.
- ✓You’ve asked an AI to analyze a screen and had no way to check what it gave you.
- ✓You’re planning a screen and want to know how many replicates and reads it needs before you spend the money.
- ✓You want to work at your own pace, on your own laptop, without a schedule to keep.
This isn’t for you if…
- ✕You already write your own screen pipelines. You’ll find this slow — it’s built for someone starting at zero.
- ✕Your screen isn’t read out by counting guides. Arrayed screens and single-cell CRISPR screens are different analyses.
- ✕You want to learn to program. You won’t, on purpose: you’ll learn to run, read and check a pipeline that’s already built.
- ✕You want someone to answer your questions as you go, or to look at your own screen with you. That’s the live workshop.
Dr. Lei Guo
Computational biologist · Founder, NGS101.com
A computational biologist with more than a decade analysing next-generation sequencing data. Hundreds of projects, more than two hundred collaborators, and a tutorial library that draws around thirty thousand visits a month.
He has trained individual researchers, whole labs and entire departments — most of them wet-lab scientists who had never opened a terminal before they started.
And he started there himself. Every course claims a decade of experience. Almost none of them can claim the person teaching it began exactly where you’re standing, got the standard advice, and lost years to it.
Same analysis.
Two different things.
The live workshop is for anyone who’d rather not face the setup and the first commands alone. Here is exactly what changes, including the parts that don’t flatter the recorded course.
| Recorded course$297 | Live workshop$597 | |
|---|---|---|
| Format | One recorded session, 2 h 35 m. Watch any time. | A live, hands-on session with me. |
| When you start | The moment you buy. | 2027, date to be announced. The waitlist hears first. |
| Setup | Your own laptop, with a step-by-step guide for Windows and Mac. | We set up your own laptop together, live, and fix whatever goes wrong. |
| Questions answered | No. It’s self-serve — you get the material, not me. | Yes, live, as you hit them. This is what the workshop sells. |
| Your own screen | No. You learn on the example, then run the pipeline yourself. | A private 1:1 where we run the pipeline on your own screen data. |
| Materials | Video, slides, handout, workbook, pipeline and example data. Yours to keep. | All of the same — plus this recorded course, on enrolment, to watch beforehand. |
Your $297 comes off the price of the live workshop. You never pay more for having started with the recording — the total is the same $597 either way.
Which makes this the low-risk way in: take the course, work through it at your own pace, and if you get to the end wanting someone to sit with you and your own screen, the workshop is still there at no penalty.
How it works: reply to your welcome email when you’d like a seat, and I’ll send you a single-use code for the full $297 off.
Want the live version instead? See the live workshop →
The questions I
actually get asked.
If yours isn’t here, it’s at the bottom of this list.
I’ve never used the command line. Is this honestly going to work for me?
That is who the course is built for. Part 1 starts at a laptop that has never run a command, installs everything with you, and teaches only the handful of commands the analysis uses. You won’t learn to program, and you don’t need to: the pipeline is already built.
The free walkthrough above is the honest test. Ten minutes, the whole course on screen. If the pace looks wrong for you, you’ll know before you spend anything.
What do I need on my computer?
A Windows or Mac laptop, and room for the download: about 7 GB — the course itself as one ZIP, plus the five raw FASTQ files of the example screen as separate downloads, because they’re large. Setup takes about an hour, most of it unattended downloading.
No cluster and no cloud account. Counting the full genome-wide example screen took about eight minutes on an ordinary laptop, and every figure draws in about half a minute.
Will the pipeline work on my own screen?
If it’s a pooled screen read out by counting guides, yes — that’s what it’s built for. It takes your own library file in whatever layout it came in, your own samples and comparisons, and mouse screens as well as human. If your core already sent you a count table, one setting skips counting entirely.
What the recorded course doesn’t include is me looking at your screen with you. If that’s what you want, it’s the live workshop, and the comparison above lays out the difference.
Can’t I just ask AI to analyze my screen?
You can, and the course shows you how to use AI well. The problem is checking what it hands back: the test above went from 103 essential genes to 3,540, lost the paper’s main finding, and never printed an error.
This course is what lets you catch that. Once you know which comparison answers your question, what the quality numbers should look like, and when a hit rests on one lucky guide, AI becomes a fast assistant instead of an unchecked one.
Can I get an invoice so I can expense it?
Yes. Email lei.guo@ngs101.com and say what the invoice needs to show — your institution’s name and address, a PO number, whatever your finance office asks for. If they need the paperwork before you can pay, say so and I’ll send it first.
If I take the live workshop later, do I pay twice?
No. The full $297 comes off the workshop — $297 here plus $300 then is the same $597 as enrolling directly, so starting with the recording never costs you more. Reply to your welcome email when you want a seat and I’ll send you the code.
What happens if I get stuck?
Straight answer: this is a self-serve course, and it doesn’t come with a support channel — no email help, no forum, no office hours. That’s why it’s $297 and the workshop is $597.
What you do have: a troubleshooting table for the errors people actually hit, the free written tutorial covering every step, and the habit the course teaches from the start — paste the exact error and the command you ran into an AI assistant. For a beginner that is genuinely the fastest route.
If having someone to ask is the thing you actually want, join the live workshop instead. That’s what you’re paying the difference for.
Still unsure? Questions about buying, access or invoices go to lei.guo@ngs101.com and I’ll answer them. For help with your own analysis, that’s what the live workshop is for.
The next time a gene list lands on your desk,
be the one who can tell whether it’s true.
One session, two and a half hours, a real published screen, and a pipeline you can point at your own. If you were looking for the catch, it’s this: you have to understand what you’re running. That’s also the point.
All sales are final — this course does not come with refunds. Every part of it, the video included, is downloadable and yours to keep the moment you have access, so a refund would mean returning nothing.
That is exactly why the full ten-minute walkthrough is free and sits above. It shows you the whole course, every deliverable and the real slides before you spend anything. Watch all of it, then decide.
And what it doesn’t include: this course is self-serve. It does not come with email support, a forum, office hours, or any other way to reach me with questions. If that is what you want, it is the live workshop.
What “yours to keep” means. Everything downloads to your own machine. Once you have the files, nothing you paid for depends on a server staying up — the video plays, the scripts run and the data is already there. Your download area on this site stays open for at least twelve months after you buy.
If you move up to the live workshop later. The full $297 comes off the workshop price. Reply to your welcome email when you want a seat.
The full terms. Everything above, plus the licence and how long access lasts, is on the Terms of Purchase page. You will be asked to agree to it at checkout.