Journal: STAR Protocols

Article Title: Protocol for 3D-guided sectioning and deep cell phenotyping via light sheet imaging and 2D spatial multiplexing

doi: 10.1016/j.xpro.2025.104296

Figure Lengend Snippet: 3D spatial and data analysis workflow 3D spatial analysis of 3D-IF stained and optically cleared samples with UltraMicroscope Blaze™ light sheet microscope (Step 16). Post processing (stitching) in case data was acquired with tile-scanning (Step 18). Surfaces generation of autofluorescence and target region with Imaris software (Oxford Instruments) and target plane definition with “Oblique Slicer” tool (Step 20 and 20).

Article Snippet: Continue the workflow in Imaris to create orientation marks for further analysis.

Techniques: Staining, Microscopy, Software

Journal: bioRxiv

Article Title: Multimodal AI for Single cfDNA Profiling and Cancer Screening

doi: 10.64898/2025.12.29.696856

Figure Lengend Snippet: cf-nucleosomes in plasma were captured with H3K4me3-affinity antibody-coated magnetic beads and subjected to an enzymatic methylation-conversion sequencing workflow that preserves multi-omic co-information. For each cfDNA fragment (read), multiple omic features were extracted by bioinformatic processing and mapped into a tailored vector-embedding representation. The multimodal representations were merged into a single cfDNA token. During model training, the embedding layer, Transformer layers, and multi-task heads are trained end-to-end. The learnable embedding layer can attend to multiple aspects of each omic modality and produce semantically rich encodings; the multimodal fusion layer and Transformer layers enable crosstalk among modalities. Model outputs include per-cfDNA scores, cfDNA gene-batch–level diagnoses, and information-rich latent states.

Article Snippet: Eluted cfDNA underwent nick repair by Taq Ligase (NEB, M0208) and dual-strand library construction, followed by enzymatic methylation conversion using a NEBNext Enzymatic Methyl-seq (EM-seq) workflow (NEB, E7120L) with unique dual indices (E7140L).

Techniques: Clinical Proteomics, Magnetic Beads, Methylation, Sequencing, Plasmid Preparation

Journal: bioRxiv

Article Title: Multimodal AI for Single cfDNA Profiling and Cancer Screening

doi: 10.64898/2025.12.29.696856

Figure Lengend Snippet: Model architecture and training workflow. (A) Expression-aware gene-wise sampling mixes target and depth-matched normal background to produce pseudo-labeled cfDNA batches; expression metrics (depth ratios and up/down direction) are concatenated with gene semantics into the CLS token and fed together with multimodal cfDNA tokens into a standard Transformer. Attention pooling aggregates token outputs to produce per-cfDNA scores, and a secondary attention pooling produces a CLS embedding for batch-level discrimination. (B–D) Training trajectories for loss and monitoring metrics; the initial random-guess baseline is annotated. Each epoch corresponds to a full pass through the training samples (∼40,000 steps per epoch, with a step defined as one batch). cfDNA enrichment is defined as the ratio of mean cancer-origin cfDNA score to mean noise cfDNA score and quantifies improvement in signal-to-noise.

Article Snippet: Eluted cfDNA underwent nick repair by Taq Ligase (NEB, M0208) and dual-strand library construction, followed by enzymatic methylation conversion using a NEBNext Enzymatic Methyl-seq (EM-seq) workflow (NEB, E7120L) with unique dual indices (E7140L).

Techniques: Expressing, Sampling, Labeling