PALYIM: A Modular, Open-Source Science Gateway for AI-Enabled Palynology Image Analysis

We introduce PALYIM (PALYnology IMage Analysis Platform), a web-based, intelligent, modular, and extensible platform under development for processing and visualizing Z-stack Whole Slide Image collections using published AI models for palynological analysis. Palynology refers to the study of pollen grains, spores, and other microscopic organic-walled materials, called palynomorphs, isolated from sediment and rock samples [2]. Palynologists traditionally count fossil pollen and spores mounted on microscope slides visually. The advent of reliable and affordable slide scanning microscopes allows us to produce virtual, digital representations of these slides, though each high-resolution image may use ~10s of GBs of storage. Despite a growing body of published work on automated identification methods, pollen identification remains a highly specialized, laborious, and primarily visual skill. There is a mismatch between the expertise of many pollen experts and the skills required for developing and implementing computer vision, deep learning, data management, and HPC workflows. Additionally, most pollen analysts will have limited access to advanced computing resources needed to implement these workflows. PALYIM aims to significantly reduce these barriers, a pattern also observed in other scientific domains that rely on high-resolution image analysis, such as histopathology. We are developing PALYIM research software modules focused on ease of use and adaptability for research groups using different microscopy data formats and research foci. We test these modules with data from three groups to avoid isolated development and release them as open source once they reach basic maturity. We present the high-level architecture of PALYIM, which includes Clowder V2 [6] at its core, and describe the developed modules, including Tile Cropping [8], Region of Interest Segmentation [9], Binary Pollen Classification for initial filtering of pollen and non-pollen segments, Pollen Clustering and Annotation for training data generation, and planned modules, including final Pollen Classification to identify different pollen morphospecies. Clowder offers customizable, extensible features, including event listeners for messaging and data processing, a frontend for data organization and visualization, user account management, HPC batch-job connectors, and a database. The PALYIM modules are command-line, container-ready applications customizable via run-time parameters, using Singularity and SLURM for HPC execution, and Docker for local runs. These will be integrated into Clowder via event listeners (planned). We share initial results and improvements, including images and metrics (e.g., Tile Cropping I/O optimizations yielded a ~3–5x speedup, cutting multi-day processing times to ~1.5 days) from applying these modules to process a subset of microscopy data obtained from the Trans Amazonian Drilling Project [1] on NCSA Delta [4]. We also outline our future work, including creating an end-to-end workflow connecting these modules, data processing of digitized microscopy data collections, training researchers in various career stages, and integration with the biological specimen image repository MorphoSource [7], the fossil specimen occurrence database, the Paleobiological Database [10], and Bioimages Archive [5] to seamlessly coordinate image analysis with the publicly accessible archival of specimen images and taxonomic information. Modules of PALYIM in its nascent form have been applied to palynology research [3]. With further development and feedback, we expect the palynology community to adopt PALYIM widely, with potential to extend to other scientific domains that use similar high-resolution image data. References [1] Baker, P.A. et al. 2015. Trans-Amazon Drilling Project (TADP): origins and evolution of the forests, climate, and hydrology of the South American tropics. Scientific Drilling. 20, (Dec. 2015), 41–49. https://doi.org/10.5194/sd-20-41-2015. [2] Erdtman, G. 1963. Palynology. Advances in Botanical Research. R.D. Preston, ed. Academic Press. 149–208. [3] Feng, J.T., Satheesan, S.P., Kong, S., Donders, T.H. and Punyasena, S.W. 2025. Addressing the “open world”: detecting and segmenting pollen on palynological slides with deep learning. Paleobiology. (Aug. 2025), 1–14. https://doi.org/10.1017/pab.2025.10059. [4] Gropp, W., Boerner, T., Bode, B. and Bauer, G. 2023. Delta: Balancing GPU Performance with Advanced System Interfaces. (2023). [5] Hartley, M., Kleywegt, G.J., Patwardhan, A., Sarkans, U., Swedlow, J.R. and Brazma, A. 2022. The BioImage Archive – Building a Home for Life-Sciences Microscopy Data. Journal of Molecular Biology. 434, 11 (June 2022), 167505. https://doi.org/10.1016/j.jmb.2022.167505. [6] Marini, L., Gutierrez-Polo, I., Kooper, R., Satheesan, S.P., Burnette, M., Lee, J., Nicholson, T., Zhao, Y. and McHenry, K. 2018. Clowder: Open Source Data Management for Long Tail Data. Proceedings of the Practice and Experience on Advanced Research Computing (New York, NY, USA, July 2018), 1–8. [7] MorphoSource: https://www.morphosource.org/. Accessed: 2026-08-05. [8] Puthanveetil Satheesan, S., Feng, J., Kong, S. and Punyasena, S. 2025. A Command Line Interface for NDPI Format Image Tile Cropping. Zenodo. [9] Puthanveetil Satheesan, S., Kong, S. and Punyasena, S. 2025. A Command Line Interface for Basic Post-Processing of Pollen Detection Images Using Segment Anything Model 2 (SAM 2). Zenodo. [10] The Paleobiology Database: https://paleobiodb.org/#/. Accessed: 2026-08-05.