A modular generalist-specialist AI framework for ROI selection across spatial profiling workflow

Castillo, S. P.; Gautam, T.; Pinao Gonzales, K. B.; Salvatierra, M. E.; Serrano, A.; Ercan, C.; Rodriguez, B. L.; Acosta, P.; Chen, P.; Shokrollahi, Y.; Lau, A.; Kwong, L. N.; Huse, J. T.; Pan, X.; Patient Mosaic Team,; Solis Soto, L. M.; Yuan, Y.

doi:10.64898/2026.06.26.734862

bioRxiv · 10.64898/2026.06.26.734862

A modular generalist-specialist AI framework for ROI selection across spatial profiling workflow

Castillo, S. P.·Gautam, T.·Pinao Gonzales, K. B.·Salvatierra, M. E.·Serrano, A.·Ercan, C.·Rodriguez, B. L.·Acosta, P.·Chen, P.·Shokrollahi, Y.·Lau, A.·Kwong, L. N.·Huse, J. T.·Pan, X.·Patient Mosaic Team,·Solis Soto, L. M.·Yuan, Y.

View original source ↗Export metadata ↓

Abstract

Selection of regions of interest (ROIs) is often a crucial step in spatial molecular profiling and many pathology tasks, with substantial implications for research reproducibility and biological interpretability. To provide a reproducible and adaptive framework for AI-guided ROI selection, we developed a modular generalist-specialist solution across spatial profiling platforms. In a cohort comprising 55 tumor types from 160 tissue donors profiled using NanoString Digital Spatial Profiling and multiplex immunofluorescence, we first established a protein-profiling reference atlas capturing compartment-specific immune, checkpoint, stromal, and proliferation patterns. We then developed an AI Specialist Task-Oriented Model for ROI Selection (ASTROS) and tested comprehensive benchmarks considering specialist-only (ASTROS), generalist-only (PLIP/GFM), and hybrid generalist-specialist strategies, showing that the latter provides a balanced tradeoff across slide-level signal preservation, pathologist-reference concordance, within-slide placement consistency, and large-slide computational efficiency. We further demonstrated the feasibility of virtual staining for ROI preview and modular ROI placement for other spatial omics technologies, Visium and Visium HD workflows. Together, these results support our proposed framework to enable ROI selection responding to unmet needs for reducing inter-rater variability, reproducibility, and versatility in spatial profiling experiments.

Explore related subjects

pathology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Castillo, S. P., Gautam, T., Pinao Gonzales, K. B., Salvatierra, M. E., Serrano, A., Ercan, C., Rodriguez, B. L., Acosta, P., Chen, P., Shokrollahi, Y., Lau, A., Kwong, L. N., Huse, J. T., Pan, X., Patient Mosaic Team,, Solis Soto, L. M., Yuan, Y.. 2026-07-01. A modular generalist-specialist AI framework for ROI selection across spatial profiling workflow. https://doi.org/10.64898/2026.06.26.734862

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Prevention of Unc13a cryptic splicing is sufficient to preserve memory

TDP-43 dysfunction is thought to underlie frontotemporal dementia and limbic-predominant age-related TDP-43 encephalopathy, neurodegenerative dementias currently without effective therapy. Therapeutic strategies are designed to correct individual cryptic targets of TDP-43, such as UNC13A, whereby its cryptic splicing compromises synaptic function, yet the sufficiency of such an approach to prevent memory deficits is unclear. Using a forebrain neuron-specific TDP-43 knockout mouse model that recapitulates TDP-43 dysfunction occurring during early stages of human disorders, we found here that prevention of cryptic splicing to include that of Unc13a attenuated memory deficits. We show that genetic ablation of Unc13a cryptic exon solely in such TDP-43 knockout mice is sufficient to preserve cognition, supporting the clinical value of targeting UNC13A to mitigate memory deficits. Prevention of cryptic splicing of multiple targets of TDP-43 additionally attenuate neuron loss. For optimal outcomes in TDP-43 related dementias, these findings thus strongly support strategies designed to repress cryptic splicing of multiple targets of TDP-43, including UNC13A.

pathology↗

bioRxiv2026

Social-Cognitive Dysregulation Model of Misophonia: Perspective from a Behavioural Study

Misophonia is increasingly conceptualized as more than a disorder of sound tolerance, with trigger over-reactivity shaped by the social meaning of sounds, inferred intentions, and representations of others actions. We tested a social-cognitive dysregulation model of misophonia in (N = 341) adults using behavioural measures of Theory of Mind and emotion recognition, alongside measures of reflective functioning, empathy, alexithymia, and mimicry. Dimensional associations with misophonia severity and its five different dimensions were examined while accounting for age, sex, sound sensitivity, and anxiety/depressive symptoms. Increased misophonia severity was associated with less accurate and slower mental-state inference and emotion recognition. ToM accuracy effects were evident for more complex, cognitive, and affective mentalizing, but not for simpler mentalizing or physical control judgments, while emotion-recognition accuracy differences emerged for positive but not negative stimuli. Greater severity was also characterized by reduced certainty and greater uncertainty about mental states, greater difficulty identifying one s own feelings, and elevated alexithymia, whereas global self-reported empathy was largely preserved. Misophonia severity further predicted a greater propensity to mimic trigger-producing actions or sounds; 41% of participants exceeding the S-Five clinical cutoff (> 87) endorsed mimicry, which was particularly associated with a subjective restoration of control. Findings remained robust following influential-case sensitivity analyses. These results reveal a selective disturbance in self-other representation spanning mentalizing, emotion decoding, emotional self-representation, and embodied regulatory processes. They position misophonia within a broader social-cognitive framework in which auditory-affective reactivity may intersect with altered inferential and sensorimotor processing, while stopping short of causal inference.

pathology↗

bioRxiv2026

Lamin A/C depletion from myofibers and satellite cells in mice reveals selective muscle pathology

Mutations in the laminA/C gene (LMNA), which encodes the nuclear lamina proteins lamin A and lamin C (lamin A/C), have been linked to different human diseases affecting different tissues. Most LMNA mutations cause cardiomyopathy and muscular dystrophy, such as autosomal dominant Emery-Dreifuss muscular dystrophy. Recent studies to understand striated muscle laminopathies have taken advantage of Lmna conditional knockout mice to examine the effects of lamin A/C depletion in cardiomyocytes and cardiac fibroblasts. However, the role of lamin A/C in skeletal muscle has largely been uncharacterized using conditional knockout mice. We used different mouse lines to deplete lamin A/C from specific cell types in striated muscle. Lamin A/C depletion from fetal myofibers and cardiomyocytes led to no observable phenotype in the skeletal muscles despite leading to dramatic heart dilation and early lethality. Depletion of lamin A/C from both skeletal myofibers and satellite cells was lethal, with the most dramatic myopathic abnormalities observed in the intrinsic muscles of the tongue. The presence of lamin A/C in skeletal muscle satellite cells prevented the development of lethal myopathy when the proteins were deleted only from differentiated myofibers. Overall, our results provide a foundation for understanding the roles of lamin A/C in muscle maintenance and development, including the variable skeletal muscle involvement and much more invariant cardiomyopathy in patients with LMNA mutations.

pathology↗