Lessons from Cellino: How to achieve end-to-end lab automation

The days of manual pipetting and tedious sample prep are on their way out. Automation is rapidly transforming modern R&D and expanding the scope of what’s possible — even the biggest, most audacious ideas.

At Cellino Biotech, we’re using automation to build a scalable biomanufacturing platform for personalized cell therapies. Conventional processes for creating autologous cell therapies, which are derived from a patient’s own cells, are nearly impossible to scale. It requires too much manual handling, it’s costly, and the results are highly variable. 

But autologous cell therapies, including induced pluripotent stem cells (iPSCs), have a huge advantage over off-the-shelf allogeneic cell therapies when it comes to risk of an adverse immune response. With iPSCs, you can reprogram a patient’s own cells into almost any cell type, laying the groundwork for regenerative medicines for some of the toughest diseases, including Parkinson’s, diabetes, and heart disease. The possibilities are endless.

That’s why our vision is to scale biomanufacturing for as many patients as possible, while also lowering production costs and accelerating manufacturing times. Automation, along with stem cell technology, machine learning, and laser physics, is how we’ll get there.

At Benchtalk, I shared how we’re tackling end-to-end automation at Cellino, with practical learnings across our experiences in physical automation, lab automation, and data automation. Here are some of the biggest takeaways. 

Michael Cheng, Automation Engineer at Cellino, spoke at Benchtalk 2024 about how his team is using automation, along with stem cell technology, machine learning, and laser physics, to scale biomanufacturing for as many patients as possible.

1. Start with a science-driven data strategy

At Cellino, we generate an incredible amount of data. Every day, we process, monitor, and analyze hundreds of different cell lines across multiple automated systems. 

In fact, our core technology depends on an automated optical bioprocess work cell, which enables us to image and manipulate cells — all while keeping them in a closed system. With automated daily image capture, we can use AI/ML to derive insights about the cell colonies over the course of several weeks. Next, our proprietary laser system precisely removes undesired cells, allowing us to isolate high quality patient-derived iPSCs with optimized cell phenotypes for downstream processes.

Our second automated system handles cell culture, capturing continuous images for high-dimensional image analysis. Here, we’re tracking growth rate, confluence, and other measures to optimize the cell culture process for iPSC clone expansion. 

Our third automated system is our quality control (QC) work cell. We use this system to run biochemical assays like qPCR, and other fluorescence- or flow-based assays.

The sheer amount of data generated by all these instruments and systems requires a robust, enterprise-wide data strategy. We started out with paper lab notebooks and manual data transfers — but that needed to change.

So we identified four key requirements:

• Structured data: Structured data, stored in a queryable form, is the foundation of it all. A unified data format, even across instruments and systems, is essential for AI- and ML-driven approaches.

• Data traceability: With so many different instruments and systems, it’s critical to have a reliable source of truth for where your data come from (e.g. time-stamped data, instrument log files, experimental run).

• Data provenance: Automated linkage of data between experimental results and the corresponding biological entity prevents data loss.

• Integration: When AI/ML models rely on large, diverse datasets, it’s vital to integrate your systems and enable creation of multimodal datasets.

We wanted a unified R&D platform that supports these foundational requirements, allowing our team to concentrate on the AI/ML aspects of our work. After reviewing a few options, we chose Benchling. 

We appreciated that Benchling’s Registry system enables structured data capture, entity lineage traceability, and data provenance — while also being user-friendly and flexible enough that we can quickly innovate without being gated by a lengthy change management process. Its various modules also enable us to conduct many aspects of our work on a unified platform, with easy data sharing, audit, and permission control. 

Today, our entire R&D team uses Benchling to centralize our data, standardize experiment documentation, track inventory, automate instrument data ingestion, and more — all in a regulatory compliant environment.

2. Automate your data capture 

By automating data capture, we’ve standardized how our experimental results are associated with individual biological samples. With Benchling Connect, we automatically ingest instrument data from our plate readers and qPCR machines, and results are immediately accessible across our R&D teams. 

Previously, we had to rely on USB drives to transfer instrument data between computers, and at times, had to manually copy and paste data across applications. By reducing these tedious, potentially error-prone tasks, our scientists no longer have to worry about data scattered across various hard drives, and ultimately, have more time for science. Reliable, standardized data ingestion that integrates seamlessly into our ELN/LIMS is a game changer.

Because Connect is native to Benchling and supports bi-directional integration, it automates tasks beyond data ingestion. For example, by simply inputting two parameters to configure a Run within an ELN, Connect can combine the input with sample information already saved in Benchling to automatically create Hamilton worklists for our liquid handlers, saving scientists time, and establishing reliable sample tracking in Benchling.  

Being able to trust the provenance of your data, and integrating results across systems, unlocks robust datasets for insight-driven decision making — both within automated workflows and by individual scientists. 

More automation begets more standardization. Don’t underestimate the power of standardizing your data, experimentation, and processes — it’s the foundation for AI- and ML-driven insights.

3. Enable sample traceability across the entire product lifecycle

With thousands of biospecimens moving between multiple systems, we needed a unified R&D platform to track sample movements, including via automated liquid handlers. Centralizing our inventory ensures sample traceability and facilitates seamless transitions across scientific teams — and across the R&D lifecycle, from biospecimen intake, through genetic reprogramming, optical bioprocess, QC, and to the final product.

For example, look at our QC workflow. When our upstream scientists submit a request to our QC team, everything is captured in Benchling — from the initial request and sample handoff, to execution of QC assays according to workflow templates. Afterwards, assay results are automatically ingested back into Benchling for analysis, triggering a notification to the original requestor.

Now, with upstream and downstream processes connected to our core bioprocess, we can evaluate the QC data, relate it back to specific samples, and use AI/ML to generate insights for improving our bioprocess. 

4. Make sure your systems are compatible — and scalable

The messy truth is that while a single software solution for all your R&D would be ideal, most companies are still using a variety of applications, including those developed in-house for your specific scientific needs. This is true for Cellino, and that’s why it was a priority to adopt a connected R&D ecosystem that could easily integrate with multiple solutions. 

Besides being compatible with each other, your systems also have to be set up in a way that’s compatible with automation. That includes software and hardware — and it has to be scalable over your R&D processes and operations. 

At Cellino, we wanted a unified platform that lets us continually expand the scope of data that we can generate and apply to our AI/ML models. For this, we found Benchling easily extensible, leveraging its open APIs, developer platform, and technical partner ecosystem.

Automation is the future, but you’ll need to prepare now

End-to-end automation must start with a strong digital data foundation. As a startup, you might think it’s too soon to start investing in a solution for your data — but in fact, the earlier you do, the better off you are.

Automation unlocks more than high-throughput experimentation; it sets the flywheel in motion for more experimental data, better AI/ML models, new insights, and ultimately, more breakthroughs. Automation is how we accelerate the future of discovery.