How RASTRUM™ Allegro and ViQi AutoHCS™ Reveal What Traditional Workflows Miss
In collaboration with Teresa Findley, PhD, Senior Scientist, Bioimaging Informatics at ViQi Inc.
At SBI2 2025, Inventia Life Science collaborator ViQi Inc. demonstrated what becomes possible when high-fidelity 3D models generated with RASTRUM™ Allegro are paired with machine-learning–driven phenotypic profiling using ViQi’s AutoHCS™ toolkit.
The result is a workflow that uncovers biological differences that are less visually obvious, rarely detected in conventional high-content datasets, and highly valuable for early discovery programs.
As 3D cell models become more widely adopted, one of the key considerations for researchers is not simply how images are acquired, but how complex 3D data will ultimately be analyzed and interpreted.
The most important question is not whether a 3D system produces more data, but whether it helps researchers see more biology.
This workflow shows that when 3D models are consistent and biologically informative, they enable a deeper view of drug activity that AI can quantify with clarity and statistical confidence.
“Consistent 3D models offer a wealth of data, but their complexity demands more than traditional analysis can provide,” says Teresa Findley, Senior Scientist, Bioimaging Informatics at ViQi Inc. “ViQi’s AI enables researchers to perform precise, high-sensitivity queries across these vast datasets, maximizing biological insights while streamlining the costs of data acquisition and storage. Together, we are making complex 3D biology both scalable and interpretable.”
In this context, value is defined by the ability to detect:
This level of insight is where the combined RASTRUM Allegro and ViQi AutoHCS workflow excels.
In the SBI2 study, reproducible U-87MG glioblastoma cultures were generated using RASTRUM and analyzed using ViQi’s AutoHCS toolkit. Images were processed using a segmentation-free approach in which each image was tiled into small regions, with approximately 2,700 image features extracted per tile. No segmentation was applied, preserving the full phenotypic complexity of each image tile.
Rather than focusing on platform-level comparisons, the key observation was that highly consistent 3D models produced unusually clear compound separation across the analysis.
“By applying ViQi’s AI technology to the standardized RASTRUM™ glioblastoma models, we achieved a remarkably clean separation of compounds based on their distinct phenotypic signatures,” says Dr. Findley. “Consistent biological inputs are very helpful in training high-performance models; they reduce background variation, allowing our algorithms to focus on the true therapeutic effect. This partnership optimizes both the precision of our findings and the speed of the entire discovery pipeline.”
These outcomes reflect improvements in biological signals driven by factors such as uniform cell distribution, consistent hydrogel formation, and reduced well-to-well variability. Together, they create datasets that support robust AI-driven analysis.
Some compounds, such as paclitaxel or cisplatin, produce strong morphological changes at high doses that can often be identified visually.
Other compounds do not. 5-fluorouracil (5-FU), for example, acts through more subtle mechanisms linked to nucleotide and DNA metabolism. In U-87MG cultures, many intermediate doses of 5-FU appear visually similar.
In the RASTRUM Allegro dataset, however, AutoHCS delivered clear compound separation and accurate dose prediction across all nine doses tested.
“Even when limited to a single z-plane, ViQi’s AI toolkits achieved clear separation between compound conditions—including subtle phenotypic shifts at intermediate doses of 5-FU. This underscores the power of combining high-fidelity 3D cultures with sensitive image analysis to deliver rapid, high-confidence results,” Findley explains. “For researchers, our system has the potential to provide a vital safety checkpoint by identifying subtle off-target effects early in the pipeline.”
The analysis yielded clear mapping across phenotypic space that could not be resolved visually, enabling a high degree of separation between test compounds, supported by:
These insights support better decision-making in early discovery by helping researchers validate hits, differentiate mechanisms of action, select appropriate doses, and map new molecules to known phenotypic classes.
AutoHCS uses a segmentation-free strategy designed to capture complex phenotypes:
This approach is particularly effective for graded phenotypes, revealing relationships and subtle differences that segmentation-based pipelines often miss.
“At ViQi, we build our bioassays with a 'biology-first' mindset,” says Findley. Recognizing that critical phenotypic information might exist in the greater landscape, our pipelines are engineered to analyze the entire image content, ensuring no potentially relevant data is lost to segmentation errors.
“We utilize feature classification specifically for these applications because it excels at capturing biological continuity. This allows us to map the 'in-between' states and unexpected phenotypes that often emerge in, for example, drug dose-response curves, providing a higher-resolution view of the drug's impact.”
Researchers gain a clearer view of how compounds behave, both across and within mechanistic classes.
During the analysis, ViQi evaluated single z-planes, sparse z-plane sampling, and maximum intensity projections. Even with reduced sampling, the RASTRUM Allegro dataset maintained strong compound separation.
This finding shows that dense z-stacks are not always required, which can reduce imaging time and data load. Importantly, the efficiency benefit supports the science rather than defining it.
Even at lower sampling rates, highly separable phenotypes were preserved, highlighting that model consistency is the primary driver of insight.
Together, these results demonstrate how pairing high-quality 3D models with analysis strategies designed for complexity creates a practical foundation for advanced phenotypic screening applications, including:
To explore the data in full, access the poster presented at SBI2 2025.
→ Download the poster
To discuss how this workflow could support your research, connect with our team.
→ Connect with us
ViQi develops AI-powered, image-based bioassays that help researchers extract actionable insight from complex biological systems. Their platforms combine machine learning with advanced microscopy to support phenotypic profiling, assay optimization, and decision-making across drug discovery and biomanufacturing workflows.
ViQi’s AutoHCS™ toolkit enables high-sensitivity analysis of advanced cell models, including 3D cultures, helping teams distinguish subtle biological effects, optimize experimental design, and accelerate discovery with greater confidence.
To learn more about ViQi and their technology, visit viqiai.com
For more information about AutoHCS™ or ViQi’s AI-driven bioassays, contact Teresa Findley, PhD, Senior Scientist, Bioimaging Informatics at ViQi Inc.