Journal: Aging (Albany NY)

Article Title: Systemic factors in young human serum influence in vitro responses of human skin and bone marrow-derived blood cells in a microphysiological co-culture system

doi: 10.18632/aging.206288

Figure Lengend Snippet: The BM model secretes age-associated proteins. The BM model was statically cultured for five weeks. After two weeks, the BM cells were treated with either young or old human serum. On culture day 35, the BM cells were harvested, and the washed cell pellet analyzed using tandem LC-IMS-MS/MS proteomics. ( A ) Log2FC and −log2( p -value) of all significantly ( p < 0.05) up (orange) or downregulated (turquoise) proteins in the BM with old serum compared to young serum. Proteins regulated in the same direction in at least 4 of 5 samples are depicted as well as either upregulated (red) or downregulated (blue). ( B ) Comparison of all regulated proteins to 2772 potentially secreted proteins according to the human protein atlas, creating an overlap of 233 proteins. ( C ) Go-Term analysis of down- (left) and up- (right) regulated overlapped proteins shown in ( B ). ( D ) Heatmap showing the log2FC of the overlapped 55 proteins in ( E ) depicting upregulated (red) and downregulated (blue) proteins with old serum. ( E ) Venn diagram showing the overlap of regulated proteins that belong to the human secretome (left) and secreted proteins that significantly change upon aging (right), resulting in 55 proteins shared between the two categories. ( F ) STRING protein network of the down- (left) and up- (right) regulated proteins from the 55 overlap proteins shown in ( E ). Expression by different BM cell types is highlighted with yellow circles (granulocytes), blue circles (progenitor cells) or violet circles (monocytes). Data were obtained from one experiment with 5 replicates.

Article Snippet: According to the Human Protein Atlas list “human secretome” (proteinatlas.org) [ ], none of these proteins are known to be secreted to potentially impact other tissues.

Techniques: Cell Culture, Tandem Mass Spectroscopy, Comparison, Expressing

Journal: bioRxiv

Article Title: BioEngine: scalable execution and adaptation of bioimage AI through agent-readable interfaces

doi: 10.64898/2026.04.19.719496

Figure Lengend Snippet: a, Collaborative fine-tuning workflow. A biologist submits microscopy images to BioEngine, which runs Cellpose-SAM to produce initial segmentations. Annotators worldwide correct those masks through a browser-based tool at BioImage.io, and each annotation cycle adds new training images and triggers GPU fine-tuning on BioEngine with one click and no local installation. The progressively improved model is published to BioImage.IO and becomes immediately available through any BioEngine deployment, extending the shared model pool for the wider community. b, Segmentation montage showing Cellpose-SAM predictions on PlantSeg Arabidopsis lateral root nuclei (Movies 1–3) from baseline through four collaborative fine-tuning cycles, alongside ground-truth. Baseline Cellpose-SAM fails to detect most large plant nuclei. As more annotated slices accumulate across cycles, the model progressively recovers boundaries and separates adjacent objects. c, F1 score (IoU ≥ 0.5) on held-out test slices (± s.d., n = 3 slices per movie) increases consistently across all three movies with each annotation cycle, rising from a mean of 0.36 at baseline to 0.71 after four cycles (1,600 cumulative training epochs). d, An AI agent builds new BioEngine applications by generating the analysis workflow, deployment manifest, and web interface from a single plain-language prompt. The deployed EM Mitochondria Analyzer accepts EM image uploads and returns instance segmentations with morphological profiles on the Lucchi++ FIB-SEM benchmark (neural tissue, 5 nm/px), illustrating how BioEngine can be rapidly extended with new analysis capabilities without manual programming.

Article Snippet: Human Protein Atlas images and ground-truth annotations for model screening were obtained from the HPA Cell Image Segmentation Dataset ( https://www.proteinatlas.org ).

Techniques: Microscopy