Researchers at the VCU Massey Comprehensive Cancer Center have developed a new algorithm called TACIT (Threshold-based Assignment of Cell Types from Multiplexed Imaging Data) that they say shortens the time needed to identify cell types from more than a month to just minutes. Details of the development of TACIT, published in Nature Communications, show that it offers a scalable method to assign cell identities based on cell-marker expression profiles that is critical for cancer diagnosis and treatment planning.

“We’re using artificial intelligence to increase efficiency and also the accuracy of diagnosis,” said Jinze Liu, PhD, a Massey researcher and professor in the VCU School of Public Health. “And as we gain more data, TACIT’s ability to increase positive patient outcomes will only multiply.”

TACIT was developed by Liu and Kevin Byrd, DDS, PhD, an assistant professor in the VCU School of Dentistry, to automate and improve the identification of cell types from spatial omics data, one of the big challenges in the field of spatial biology. Spatial biology technologies allow scientists to map the location and interactions of cells in tissues with high resolution, but identifying individual cell types from these complex data has remained slow and error prone.

Traditional unsupervised clustering methods, such as Louvain algorithms used in single-cell RNA sequencing, have limitations when applied to spatial omics data, particularly when marker sets are small or when rare cell types are involved. “This sparse marker set, often of only one modality, lacks the power to separate expected cell populations in the embedded feature space, posing a formidable obstacle for unsupervised clustering to detect all cell types—especially rare ones,” the researchers wrote.

TACIT overcomes these challenges by focusing on relevant features of predefined cell types and de-convoluting ambiguous or mixed populations using unbiased thresholding. According to the researchers, “TACIT automates cell type annotation, mimicking manual gating with enhanced scalability and precision. This method excels in phenotyping based on multiplex panel design, effectively identifying both dominant and rare cell populations without bias.”

In benchmarking tests, TACIT was applied to more than 5 million cells across 51 cell types in tissues including the brain, intestine, and salivary glands. The VCU team showed that it outperformed three existing unsupervised algorithms in both accuracy and scalability. The researchers showed that TACIT maintained 81% agreement between protein and RNA data types, making it a strong candidate for translational applications.

Liu noted that TACIT could become a valuable tool for more efficiently recruiting patients for clinical trials. “One of our goals as scientists is to identify good spatial biomarkers for clinical trials so that we can predict patient responses to the trial before they even are enrolled,” he said. “We have already been working with multiple principal investigators on [VCU’s] campus to include spatial biology into clinical trials, and TACIT can provide that guidance so that we can make sure the clinical trial patients are receiving the best possible treatments.”

The algorithm’s versatility also allows it to function across multiple spatial biology platforms, with potential applications in pharmacology. “We have a large repository of [FDA] approved drugs we can map onto the tissue samples,” said Byrd. “Imagine if you could tell a patient, ‘Here’s an already [FDA] approved drug.’”

TACIT’s development builds on earlier work in single-cell omics and label transfer methods but offers a higher degree of scalability and cross-platform consistency. In one experiment, the researchers applied TACIT to the Xenium platform dataset and were able to refine cell type annotations and discover cell type-specific markers, further validating its effectiveness.

Looking ahead, the researchers plan to expand TACIT’s capabilities by integrating additional spatial omics technologies. “The combined analysis of spatial multiomics datasets in (graft-versus-host disease) revealed the importance of integrating spatial transcriptomics and proteomics for deep phenotyping,” the researchers wrote.

Despite its strengths, TACIT’s current limitations include variability in marker quality and the need for better-designed multimodal panels. “The discrepancies in our final dataset, which combines transcriptomics and proteomics on a single slide, highlight the need for better-designed multimodal panels to accurately identify cell types in spatial datasets,” the researchers wrote.

Nonetheless, Liu and Byrd view TACIT as a foundational tool for both clinical research and patient care, noting that the algorithm has the potential to integrate molecular data across platforms and cell types to support a more refined, personalized approach to cancer treatment.