The one-week automated genome-wide optical pooled screen

The paper introduces OttoSeq, an automated optical pooled screen platform that integrates fluid handling and analysis pipelines to complete a genome-wide cell painting screen in just eight days, successfully profiling over 5 million cells across more than 21,000 gene knockouts.

The Gist

Imagine you are a detective trying to solve a massive mystery: What does every single gene in a human cell actually do?

In the past, solving this mystery was like trying to read a library of millions of books, one by one, in a dark room, using a magnifying glass that required you to manually turn the pages. It took months, required a team of experts, and was incredibly expensive. This is what scientists call an "Optical Pooled Screen" (OPS). It involves taking thousands of cells, turning off specific genes (like removing a part from a machine to see what breaks), and then taking high-definition photos to see how the cell changes shape.

The problem was that the "reading" part—figuring out which gene was turned off in which cell—was slow, manual, and prone to human error.

Enter "OttoSeq": The Automated Robot Librarian

This paper introduces a new system called OttoSeq. Think of it as building a robot librarian that can not only read the books but also organize the entire library in a single week.

Here is how they did it, broken down into three simple parts:

1. The Robot Arm (Otto2)

Previously, scientists had to stand over a microscope for hours, manually pipetting liquids (tiny droplets of chemicals) onto the cells to "read" their genetic codes. It was like trying to paint a masterpiece with a toothbrush while someone kept shaking the table.

The team built Otto2, a robotic fluid-handling system.

• The Analogy: Imagine a high-speed, automated coffee machine that doesn't just brew one cup, but can brew, clean, and brew again for 12 different cups simultaneously, without ever spilling a drop.
• The Result: Instead of a human spending 90 minutes per cycle, the robot does it in about 10 minutes. It works tirelessly, day and night, without getting tired or making mistakes.

2. The Fast-Forward Button (Brieflow)

Once the robot takes the photos, you have millions of images to analyze. Usually, this requires a computer scientist to write custom code for every single experiment, which is like hiring a different architect to design a house for every single brick.

The team used Brieflow, a modular software pipeline.

• The Analogy: Think of Brieflow as a set of Lego blocks. Instead of building a new machine from scratch, you just snap together pre-made blocks (one for finding cells, one for reading barcodes, one for grouping them). If you need to change something, you just swap one block out without breaking the whole structure.
• The Result: The software could adapt instantly to the robot's speed, processing data in real-time rather than waiting weeks.

3. The AI Translator (MozzareLLM)

After the computer groups the cells by how they look, scientists usually have to spend months reading scientific papers to figure out why those cells look that way. "Oh, this group of cells looks weird; let me read 50 papers to guess which biological pathway is broken."

They used MozzareLLM, an Artificial Intelligence tool.

• The Analogy: Imagine you have a pile of 320 different puzzle pieces. Instead of a human trying to guess the picture, you hand the pile to a super-smart AI that has read every book in the library. The AI instantly says, "These 320 pieces all fit together to make a picture of a 'Garbage Disposal Unit' (the cell's waste system)."
• The Result: The AI summarized the biological meaning of the data in hours, not months.

The Grand Experiment

The team put all these tools together to run a genome-wide screen.

• The Scale: They tested 21,732 different genes (almost every gene in the human genome).
• The Speed: They did it in 8 days.
• The Volume: They looked at 5 million cells.

In the past, this same experiment would have taken a team of experts many months of hard labor. With OttoSeq, it was done in less than a week, mostly by machines working autonomously.

Why This Matters

This isn't just about being faster; it's about democratizing discovery.

• Before: Only a few elite labs with huge budgets and specialized experts could do this kind of deep-dive research.
• Now: Because the process is automated, cheaper, and faster, regular researchers in hospitals and universities can now ask these big questions.

The Bottom Line:
The authors turned a slow, manual, "craftsman" process into a fast, automated, "factory" process. They didn't just build a better microscope; they built a factory that can automatically discover how our bodies work, potentially leading to new cures for diseases much faster than ever before.

Technical Summary

1. Problem Statement

Optical Pooled Screens (OPS) represent a powerful method for high-dimensional, single-cell phenotyping of CRISPR perturbation libraries by reading genetic barcodes via in situ sequencing (ISS) directly in fixed cells. While OPS offers superior spatial and morphological fidelity compared to NGS-based methods (e.g., Perturb-seq), its adoption has been limited to specialist laboratories due to two critical bottlenecks:

• Labor-Intensive Workflow: Traditional in situ sequencing requires 1–2 hours of manual fluid handling per cycle per plate. A full screen typically demands 10+ cycles, resulting in weeks of expert hands-on labor.
• Computational Rigidity: Existing analysis pipelines are often project-specific, lack modularity, and require expert customization. They struggle to keep pace with high-throughput data generation, leading to months of analysis time.

Consequently, a genome-wide OPS has historically required many months of total turnaround time, hindering its routine use in biomedical research.

2. Methodology: The OttoSeq Platform

The authors present OttoSeq, an integrated system designed to automate both the wet-lab and dry-lab components of OPS, reducing the turnaround time from months to days.

A. Hardware: Otto2 Automated Fluidics

• Design: Otto2 is a low-cost, high-fidelity positive displacement pump system built from off-the-shelf components (peristaltic pumps, multi-selector valves, Arduino microcontroller).
• Integration: It is mounted directly onto a Cephla Squid+ fluorescence microscope. A custom heating box maintains plates at 45°C for chemistry operations, with fluidic lines threaded directly into the wells.
• Operation: The system uses a 3D-printed manifold to interface with six-well plates. It performs semi-parallelized dispensing and aspiration.
• Efficiency: Otto2 replaces ~90 minutes of manual labor per cycle with ~10 minutes of hands-on setup, enabling near-unattended, continuous operation. It supports 12 cycles of in situ sequencing.

B. Software: Brieflow & MozzareLLM

• Brieflow: A modular, open-source analysis pipeline. Its architecture allows independent processing steps (segmentation, feature extraction, barcode calling, spatial merging, clustering) to be swapped or updated without breaking the workflow.
  • Adaptations: During the screen, the pipeline was updated in real-time to handle unexpected plate position offsets (rotation-aware alignment) and to filter for bootstrap-significant perturbations to manage data scale.
• MozzareLLM: An AI-driven tool utilizing Large Language Models (LLMs) to interpret biological data. It automatically generates biological summaries, assigns pathway confidence tiers, and identifies novel gene functions from cluster data within hours, replacing weeks of manual literature curation.

C. Experimental Design

• Library: A genome-wide CROPseq-multi CRISPR-KO library targeting 20,553 genes (plus controls) with 41,906 dual-guide constructs.
• Cell Line: HeLa-TetR-Cas9 cells.
• Phenotyping: "Cell Painting" assay using four spectral channels (Hoechst, COX4, Phalloidin/AGP, Concanavalin A) at 20× magnification.
• Sequencing: 12 cycles of in situ sequencing at 4× magnification using diluted incorporation mixes to reduce cost.

3. Key Contributions

1. Automation of In Situ Sequencing: The development of Otto2, which successfully automates the fluid handling of ISS, reducing hands-on time by ~90% per cycle and enabling continuous, unattended operation.
2. Modular Analysis Pipeline: The demonstration of Brieflow's ability to adapt dynamically to hardware artifacts (e.g., plate drift) and scale to genome-wide data without requiring a complete rewrite of the codebase.
3. AI-Driven Interpretation: The integration of MozzareLLM to automate the biological interpretation of complex clustering data, significantly accelerating the discovery phase.
4. Speed Record: The successful execution of a full genome-wide screen (21,732 perturbations) in eight days, a timeframe previously impossible for this scale of OPS.

4. Results

• Throughput & Quality:
  • The screen processed 5,198,240 mapped cells across 21,732 gene knockouts.
  • Achieved a median coverage of 224 cells per gene, confirming sufficient statistical power for genome-wide morphological profiling.
  • Extracted 1,670 morphological features per cell.
• Timeline:
  • Total Turnaround: 8 days from library transduction to biological interpretation.
  • Compute Time: Approximately 31 hours of total runtime on a high-performance cloud instance (GCP c4d-standard-384), including 1 hour 41 minutes for MozzareLLM interpretation of 320 clusters.
• Biological Findings:
  • Identified 3,715 perturbations with significant morphological effects.
  • Organized these into 320 functional gene clusters.
  • Validation: High-confidence clusters (610 genes) successfully recovered known cellular machinery (e.g., ubiquitin-proteasome degradation, RNA Pol II transcription).
  • Discovery: Identified novel functional contexts, such as UBE3A (an Angelman syndrome-associated E3 ligase) clustering with core Pol II machinery, and genes not previously assayed in HeLa essentiality screens.

5. Significance

This work fundamentally shifts the accessibility of Optical Pooled Screening. By solving the bottlenecks of manual fluid handling and rigid data analysis, OttoSeq democratizes high-content, genome-wide CRISPR screening.

• Scalability: It transforms OPS from a months-long, specialist-only project into an eight-day, routine experiment.
• Cost Reduction: Automation drastically reduces labor costs and reagent waste (via optimized volumes).
• Future Impact: The platform suggests a future where high-speed, low-cost perturbation screening is accessible to a broad range of biomedical researchers, accelerating the discovery of gene function and disease mechanisms. The integration of LLMs for immediate biological interpretation further bridges the gap between raw data and scientific insight.