BioLizard wins best poster award at the European Society for Spatial Biology Conference

BioLizard is proud to have received the best poster award at the European Society for Spatial Biology (ESSB) conference, recognising our work on benchmarking cell segmentation methods for single-cell spatial transcriptomics.

Presented by Hugo Steenberghen, Michiel Van der Haegen, Alexander Scheer and Viacheslav Mylka, the poster explored how four cell segmentation methods perform across human lung and cervix cancer Xenium datasets. The team compared key quality metrics, including cell-type purity, transcript assignment, tissue coverage, and cell boundary accuracy, to better understand how segmentation choices influence downstream analyses.

Accurate cell segmentation is essential for generating reliable spatial transcriptomics data. The findings highlight how different methods involve different trade-offs, reinforcing the importance of selecting and evaluating segmentation approaches based on the dataset and research question.

The award is a great recognition of work that resonates with the spatial biology community. We’re proud of our team and grateful for the opportunity to share our research, exchange ideas, and contribute to advancing reliable, biologically meaningful insights from complex spatial data.

Explore the key findings and takeaways from our poster below.

Conclusions

• Default Xenium multimodal cell segmentation: good out-of-the-box option with a balanced number of cells and marker purity.
• Cellpose: best generalist – strong performer across coverage, purity, and transcript capture if IF is high quality and coverage of cell types is good.
• Baysor: highest transcript assignment – tends to over-segment, but yields higher and more specific within-cell transcript assignments.
• ProSeg: best cluster separation – increases cluster separation, but tends to deviate from IF boundaries.
• Future perspectives: using methods like ovrl.py and SPLIT might open the door to further reducing transcript contamination through deconvolution.