The Center for Data Innovation recently spoke with Willem Westra, PhD, Vice President of Business Development and Marketing at ThinkCyte, a Japan-based company using AI to advance biological research and develop new treatments. Westra explained how ThinkCyte is applying machine learning to cytometry, a technique for analyzing the physical and structural characteristics of cells, to help scientists identify and study cells faster and with greater precision.

David Kertai: What gap in existing cell analysis methods is ThinkCyte addressing?

Willem Westra: ThinkCyte began in a University of Tokyo lab, where our founder, Professor Sadao Ota, was trying to improve how researchers identify and sort cells. At the time, two main tools were commonly used: microscopy and flow cytometry. Each had varying limitations.

Microscopy lets you see rich structural details of cells: their shape, texture, and any abnormalities. Doctors often rely on it when making complex diagnoses, but it’s slow, manual, and not scalable, you have to examine cells one by one. Flow cytometry, by contrast, is fast and scalable. It passes cells quickly through a laser and detects fluorescent signals from chemical tags that are attached to specific proteins. The problem is that you must know in advance what you’re looking for, so you can apply the right tag. If something unusual is present but untagged, you will miss it entirely. Also, flow cytometry doesn’t give much insight into the cell’s shape or structure, it just tells you whether a tag is present.

Ota wanted a way to get structural information at high speed, without needing labels or manual inspection. That’s what led to Ghost Cytometry, ThinkCyte’s core technology. Ghost Cytometry measures how each cell interacts with light as it flows past a laser. These patterns are subtle, but carry information about the cell’s morphology, shape and structure, much like a fingerprint. These profiles aren’t images, they’re data rich, high-dimensional optical waveforms that reflect cell morphology in detail.

Kertai: What is the role of AI?

Westra: Each cell produces thousands of data points, and we analyze thousands of cells every second. Built-in AI algorithms process this data in real time, classifying cells and deciding whether to sort them based on complex traits, or phenotypes. This enables detection of subtle differences that traditional methods often miss. Essentially, users can train the AI inside the VisionSort instrument to recognize patterns invisible to the human eye, without relying on labels or prior assumptions.

Kertai: How do you address rare or underrepresented cell types in your datasets?

Westra: Detecting rare cells is one of our platform’s key strengths. At up to 3,000 cells per second, we can spot very rare cells, giving researchers enough data to study even the most unique cell subtypes in detail. For extremely rare populations, researchers may use pre-enrichment methods to concentrate target cells before analysis. Additionally, VisionSort can physically sort and collect cells, allowing for further downstream study. The combination of fast analysis, AI-driven classification, and flexible experimental design makes it possible to capture and analyze rare, underrepresented cell types with speed and precision.

Kertai: What challenges have you faced in driving data innovation in life sciences?

Westra: A major challenge has been introducing an entirely new kind of biological data to researchers. Most are familiar with DNA, proteins, or gene expression data, but we provide structured high-content morphological data that describes a cell’s physical form. Educating scientists about what this data type is and how it connects to biological function has taken time.

The second challenge has been infrastructure. Fields like genomics already have mature tools and workflows. For our unique data type, we had to build those systems from the ground up. That’s meant collaborating closely with partners and experts to develop custom software, data pipelines, and computing systems to support large-scale research.

Kertai: How could advances in AI modeling accelerate ThinkCyte’s growth or unlock new use cases?

Westra: We’re constantly evolving the way we integrate AI development in the platform. We intentionally started with relatively simple algorithms and machine learning models that could process morphology data very quickly, but we’re exploring more advanced techniques like deep learning and neural networks. These can process larger, more complex datasets and extract deeper insights as we, and our users, generate more data.

One of our objectives is to build a comprehensive database of morphology readouts that the broader research community can use to drive discovery. We’re also working to integrate our morphological data with other routinely measured biological data types, such as gene expression, DNA sequences, and proteins, known collectively as multi-omics. By combining these data layers, we aim to give researchers a more comprehensive view of cell behavior. This could lead to analysis that allows for earlier disease detection, the development of more personalized treatments, and a better understanding of how therapies affect different cell types, resulting in more effective medicines.

Source link