Cell line development teams lose weeks aligning data from instruments, assays, and imaging systems before they can even compare candidates. Clone IDs live in one system, titer data in another, morphology images somewhere else. By the time scientists piece it together, the decision window has closed.
The Tetra Lead Clone Selection Assistant connects that data automatically. Assay results, process metadata, and imaging analysis, harmonized by clone ID, passage, and time point, in one environment. Scientists compare productivity, stability, growth, and morphology side by side, then use predictive models trained on early-stage signals to forecast fed-batch performance before committing to scale-up.
Built with NVIDIA VISTA-2D for cell segmentation and AI-augmented for titer prediction, the app turns what used to take 8 months into a 2.5-month process. Teams we've worked with report 70% faster clone selection decisions and 85% lower cost of goods through higher-yield candidates and shorter timelines.
Watch the demo to see how connected data changes what's possible in cell line development.
Video Transcript below:
Lead Clone Selection is the process of identifying the cell line with the right productivity, stability and scalability to advance into biologics manufacturing.
Historically, the data that is needed to make that decision is fragmented across systems, spreadsheets and even proprietary file formats. So, that means scientists are having to spend a very significant amount of time aligning the data by clone IDs, time points, assay results, and process metadata before they can even compare the candidate that they would want to move forward.
What we're doing with the Lead Clone Selection Assistant is bringing all of these data together in one governed environment. So that means instrument and assay data are captured. They're harmonized into standard schemas and then linked with key metadata such as clone ID, passage, time point, fee strategy, and assay condition. And we built this application to demonstrate how connected clone selection data can support faster, more confident decision making.
And what we've seen is a measurable impact with a seventy percent acceleration in time to clone selection decision, an eighty five percent reduction in cost of goods through higher productivity and shorter timelines, as well as the ability to assess clone stability and productivity in parallel rather than as separate sequential activities. So in the demo that I'm gonna show you right now, I'll show how scientists can compare clone performance across productivity, growth, stability, morphology, and process conditions in a single view. So I have pulled up the Lead Clone Selection Assistant.
I've chosen the product that I'm interested in reviewing.
And now you can see that I can move through the key stages of the lead clone selection workflow at the top here starting from initial transaction.
I can look at mini pool performance if that's part of the standard workflow and also cloning, which is what I'm gonna focus on today.
So this is really where teams are deciding which clones are strong enough to advance. And this first view here shows clone performance over time, so I can look at things like titer, viability, or even viable cell density, and I can compare different clones. So right now, we're looking at the top three clones, but I could also look at top twenty, top forty eight, or all of the potential clones that I have. You can imagine the more clones that we're looking at, the little bit this graph becomes a bit more complex, but that capability is available. And then over here on the right, we have a radar plot to give a quick multidimensional comparison of top clones across several different variables at once.
And that helps show whether a clone is well rounded or might be strong in one area but weaker in another. And I think the important point here is that this type of comparison normally requires significant manual effort. So scientists are having to pull data from multiple systems, align those results by clone and time point, clean that data, and then rebuild plots for each round of analysis that they're doing.
But here, all of that work is being handled behind the scenes. So the assay data, the process data, and clone metadata are automatically linked and harmonized so that these views can be generated in just seconds instead of days.
And then from there, we can actually go deeper into clone quality.
So titer and viability are important, but they don't always tell the full story. So you might have a clone that looks productive early on, but might show signs of instability or stress later or it might scale poorly later.
And that's where morphology becomes really valuable. So in this workflow, we've brought imaging data into the same environment as assay and process data and we're using NVIDIA's VISTA 2D model for cell segmentation and the application takes that information to quantify morphology features such as circularity, cell diameter and other indicators of cell health and then align those results to clone ID, time point and condition.
Some of these metrics you might be able to get from a cell counter but where the embedding of this model becomes really powerful is looking at the clump analysis from these images. So you can get an idea of the clumping rate, the average clump size for the different candidates that you're interesting. You can also view the clump size distribution for a particular entity or across entities on a particular day.
In addition, we have other distribution analyses of other parameters that you can view.
That makes imaging analytically useful, not just something scientists are visually inspecting in isolation.
So now they can confirm whether a top performing clone is also gonna maintain healthy morphology over time, and this gives data scientists the ability to use morphology features as inputs for predictive models.
And so that brings us to the final piece of this, which is the prediction.
So here, we're using early stage data. So things like static titer, the morphology features we just talked about, flow cytometry results, ddPCR-based light and heavy chain ratios to predict final fed batch titer. So these are signals that already exist in the workflow, but they're usually too fragmented to combine in any reliable way. And so by bringing all of them together, the application helps scientists prioritize clones based not only on the performance indicators here and the health of the cells over time, but also on their projected performance at scale. And the result is faster selection, fewer waste experiments, and more confident advancement decisions.
To summarize, we've gone from siloed metrics to a fully integrated decision support system, from days of manual analysis to real time insights, and then from trial and error to predictive multidimensional clone selection. And everything you've seen is built on a unified data foundation that brings together data from all of the varied sources on the left of this graphic, and this powers the leaf clone selection assistant, the VISTA 2D model, as well as the titer prediction model that we have hosted in Databricks.
