hydrop  (GORYO Chemical)

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , Comparison between bead barcoding chemistries of HyDrop v1 and v2. b , The HyDrop v1 protocol was optimized in two ways: (1) Bead chemistry was adjusted in several optimization rounds, and (2) the barcoding of the beads was changed from a cyclic polymerase extension to a linear ligation-based chemistry. c , The design of the microfluidic chip to generate HyDrop hydrogel beads (left), alongside the image of the polymerized hydrogel bead and fluorescence image of the bead labeled with complement of Tn5 fragment capture site. d, Sanger sequencing of an exemplary HyDrop v1 bead (top) showing impurity in barcode one and barcode two and a HyDrop v2 bead with pure barcode signals across all three barcodes. e , Mouse cortex data were partly downloaded from the official 10x genomics website and partly generated in-house. Overview of the sample generation: dissection of mouse motor cortex, tissue lysis followed by droplet encapsulation for cell barcoding. For HyDrop v1 only one bead batch was generated (data reused from De Rop et al., 2022) while for HyDrop v2 seven bead batches were generated (seen in c). f , FRIP of experiments pooled for seven different bead batches of HyDrop v2 compared to pooled 10x v2, 10x v1, and HyDrop v1 samples. The median is indicated per sample in white. g, Estimation of generation costs of 110k cells in euros excluding sequencing costs. HyDrop v1: 1,398.29 euros, HyDrop v2: 1,080.60 euros, 10x v2: 15,330.70 euros. h , t -distributed stochastic neighbor embedding (tSNE) of all 110,547 cells generated with HyDrop v2 (35 experiments) colored by bead batch (seen in c), no batch correction. Data points were randomly shuffled before plotting. QC: Quality control, FRIP: Fraction of reads in peaks.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Comparison, Ligation, Fluorescence, Labeling, Sequencing, Generated, Dissection, Lysis, Encapsulation, Control

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , Characterization of bead quality and stability with different monomer compositions. Different percentages of total monomer and cross linker composition is varied during different bead productions and the stability of the beads are evaluated using three different characteristics: i.) bead size variations, ii.) bead stability post refrigeration, and iii.) dissolution time of the beads in a reducing environment relative to 10X genomics NextGEM 3’ RNA beads. b, log-transformed number of unique fragments of experiments pooled for seven different bead batches of HyDrop v2 compared to pooled 10x v2, 10xv1, and HyDrop v1 samples. The median is shown as white dot. c , Barnyard of HyDrop v2 protocol with mouse and human cell line. d. Table illustrating different parameter and characteristics of different bead production batches. e. Bright field image of the beads showing freezing artifact in suboptimal bead batch productions. f. Barcode ranking plot showing noisy barcode contamination in the HyDrop v1 beads arising from the barcode bleeding during the production. g , Fragment size plot of 10x v1, 10x v2, HyDrop v1, and HyDrop v2 experiment. 10x data shown is downloaded from 10x Genomics website.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Dissolution, Transformation Assay

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , t -distributed stochastic neighbor embedding (tSNE) of all 141,010 cells across 44 experiments colored by technique used, left: no batch correction, right: batch corrected for the used technique. Data points were randomly shuffled before plotting. b , Sequencing efficiency compared across 10x v1 and v2, multi-ome (black), HyDrop v1 and HyDrop v2 beads (7 batches). The colors are shown in a. c , t -distributed stochastic neighbor embedding (tSNE) showing the distribution of cells in combined embedding in a, 2 experiments 10x v1 with a total of 10,177 cells, 2 experiments 10x v2 with a total of 13,123 cells, 5 experiments HyDrop v1 with a total of 7,163 cells, 35 experiments HyDrop v2 with a total of 110,547 cells). d , Heatmap for top 1000 DARs per cell type correlated between 10x v2 and HyDrop v1 data. e , Scatterplot of normalized enrichment score (NES) of cell-type-specific transcription motifs, dots represent a common motif between HyDrop v1 and 10x v2 data colored per cell type. f , NES of microglia, Pvalb, and astrocytes divided per technique. g , Stacked bar plot showing the fraction of reads across each step of data processing using the PUMATAC pipeline (‘Unique, in cells, in peaks’: final fraction of sequencing reads retained in count matrices). The colors at the bottom of the plot indicate the technique of the experiment. The data is down-sampled to 36kRPC.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Sequencing

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a, Overview of study design. Drosophila melanogaster embryos were collected 16-20h after egg laying. In-house experiments were performed with 10x v2 chemistry on the 10x Chromium device, while HyDrop v2 beads were used for the open-source data generation. sciATAC-seq3 data (fragment files for the plate-based essay) were retrieved from Calderon et al., (2022). b , t -distributed stochastic neighbor embedding (tSNE) of all 607,340 cells generated with HyDrop v2 (340,604, 18 experiments) and 10x v2 (266,736 cells, 5 experiments) colored by cell type, batch correction according to wet lab protocol used to generate data. Data points were randomly shuffled before plotting. c, Violin plots of quality measurements (TSS enrichment, FRIP, log fragment counts) compared between 10x v2, sciATAC-seq3, and HyDrop v2 data . d, Genome tracks (upper) and cut site insertion of Tn5 (lower) of cell-type specific DARs down sampled to the smallest count present for each technique, 1,861 cells in neuronal cells and 8,916 cells in somatic muscle cells. The shown genome tracks are normalized for the fragment count. e , Carrot plot showing the regions of accessible chromatin ±0.5 kb around the center of DARs sorted by the highest (blue) to lowest (red) accessibility for neuronal cells and somatic muscle cells in 10x v2, HyDrop v2, and sciATAC data. DARs are calculated on fragment coverage (left) and Tn5 cut sites (right) for each technique. f , Heatmap of percental overlap of top 1,000 DARs per cell type of 10x v2 and HyDrop v2 data. g, Scatterplot of normalized enrichment score of cell-type-specific transcription motifs, dots represent a common motif between HyDrop v2 and 10x v2 data colored per cell type as shown in a, the labels are shown for the top three motifs per cell type. DAR: differentially accessible region, TSS: Transcription start side, FRIP: fraction of reads in peaks.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Generated

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a, t -distributed stochastic neighbor embedding (tSNE) of all 141,010 cells across 44 experiments (2 experiments 10x v1 with a total of 10,177 cells, 2 experiments 10x v2 with a total of 13,123 cells, 5 experiments HyDrop v1 with a total of 7,163 cells, 35 experiments HyDrop v2 with a total of 110,547 cells) colored by cell type, batch corrected for the used technique. Data points were randomly shuffled before plotting. b , Genome tracks of cell-type specific DARs down sampled to the lowest cell number present for each technique, 985 cells in oligodendrocytes, 469 cells in L2 3 IT neurons, 509 cells in microglia, and 998 cells in astrocytes. The shown genome tracks are normalized for the fragment count. c , carrot plot showing the regions of accessible chromatin ±1 kb around the center of DARs (logFC ≥ 1.5), sorted by the highest (blue) to lowest (red) accessibility for oligodendrocytes, microglia, and L2 3 IT neurons across the four different techniques. d , Heatmap of percental overlap of top 1000 DARs per cell type of 10x v2 and HyDrop v2 data. e , Scatterplot of normalized enrichment score of cell-type-specific transcription motifs, dots represent a common motif between HyDrop v2 and 10x v2 data colored per cell type as shown in a. DAR: differentially accessible region, NES: normalized enrichment score.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques:

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , t -distributed stochastic neighbor embedding (tSNE) of all 110,547 cells generated with HyDrop v2 (35 experiments) colored by cell type, no batch correction. Data points were randomly shuffled before plotting. b, Model comparison of sequence models trained on HyDrop v2 data (orange) and the BICCN (10x) public data set (i.e., 10x multi-ome data in blue). The accuracy of the predictions evaluated based on the ground truth of region accessibility is compared on class-wise (top) and region-wise level (bottom) for both the full test set in the data and cell type-specific test sets. c , Genome tracks of cell-type specific chromatin accessibility of the FIRE enhancer (mm10 chr18:61108475-61108975), regulating Csf1r expression in microglia. d , Nucleotide contribution score of the FIRE enhancer from a sequence model trained on HyDrop v2 data (top) and a model trained on 10x data. Previously described TF binding sites are highlighted for the corresponding TF. e, Prediction by the HyDrop v2 sequence model on HyDrop v2 data of chromatin accessibility for L6 CT (top) around the Chsy3 gene locus compared to actual chromatin accessibility plotted from the corresponding bigwig file (bottom). f , Specificity prediction for in vivo enhancer activity of validated mouse cortex enhancers by Ben-Simon et al. (2024) shown for the HyDrop v2 and BICCN (10x-based) model. g, Heatmap of enhancer indicated by Ben-Simon et al. (2024) identified as cell-type specific enhancer of the corresponding cell type by the HyDrop v2 model. h , i, Evaluation of motif importance scores of highly relevant regions for cell type identity indicated by the BICCN (10x) data-based model and HyDrop v2 data-based model for microglia ( h ) and astrocytes ( i ) with motifs shown in d and j, respectively, are identified and highlighted. j, Multiscale footprint of the region (mm10 chr2:38111566-38112137) of HyDrop v2 data only. Bottom tracks show Tn5 insertion, the predicted Tn5 bias and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Generated, Comparison, Sequencing, Expressing, Binding Assay, In Vivo, Activity Assay

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , Heatmap of correlation of prediction by HyDrop v2 model with ground truth (left), self-correlation of HyDrop v2 data as ground truth (right) for HyDrop v2 data. b , Heatmap of correlation of prediction by 10x v2 model with ground truth (right), self-correlation of 10x v2 data as ground truth (left) for 10x v2 data. c , Multiscale footprint of the region (mm10 chr3:110093884-110094329) in L5 ET neurons from HyDrop v2 data. Bottom tracks show Tn5 insertion, the predicted Tn5 bias and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF. d , Multiscale footprint of the region (mm10 chr3:110093884-110094329) in L5 ET neurons from 10x data. e, Multiscale footprint of the region (mm10 chr3:110093884-110094329) in astrocytes from 10x data (HyDrop v2 shown in . f , Multiscale footprint of the region (mm10 chr3:125898865-125899237) in oligodendrocytes from HyDrop v2 data (top) and 10x data (bottom) e , Multiscale footprint of the regions shown in c (upper) and d (bottom) in BICCN (10x multi-ome data). g-j , Evaluation of motif importance scores (threshold set on 4.25) overlapping regions (present in HyDrop v2 and 10x data) for cell type identity indicated by the BICCN (10x) data-based model and HyDrp v2 data-based model for oligodendrocytes ( g ), L2 3 ( h ), L5 ET ( i ), and L6 IT neurons ( j ), k-m , Mouse-specific Tn5 bias from Hu et al., (2025) plotted on cell type-specific DARs and carrot plot showing the regions of accessible chromatin ±0.5 kb around the center of DARs sorted by the highest (blue) to lowest (red) accessibility in for HyDrop v2, HyDrop v1, 10x v2, and 10x v1 data for oligodendrocytes ( k ), microglia ( l ), and L2 3 neurons ( m ).

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Sequencing, Binding Assay

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: ROC and PR curves from specificity scores per cell type on in vivo validated of the Hydrop v2 CREsted model, HyDrop scATAC-seq data and peak-scaled HyDrop v2 scATAC-seq data ( a) , and of the 10x CREsted model, 10x scATAC-seq data and peak-scaled 10x scATAC-seq data ( b ).

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: In Vivo

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a, t -distributed stochastic neighbor embedding (tSNE) of all 607,330 cells generated with HyDrop v2 (340,604, 18 experiments) and 10x v2 (266,736 cells, 5 experiments) colored by technique, batch correction according to wet lab protocol used to generate data. Data points were randomly shuffled before plotting. b, t -distributed stochastic neighbor embedding (tSNE) of 61,480 cells of 16-20h AEL extracted from sciATAC embryo atlas (Calderon et al., 2022) colored by cell type. Data points were randomly shuffled before plotting. c, Fragment size plot of HyDrop v2, 10x v2 and sciATAC-seq3 experiments. d, Genome tracks of cell-type specific DARs down sampled to the smallest count present for each technique, 1,247 cells in glia and 1,002 cells in hemocytes. The shown genome tracks are normalized for the fragment count. d, Carrot plot showing the regions of accessible chromatin ±0.5 kb around the center of DARs sorted by the highest (blue) to lowest (red) accessibility for glia and hemocytes for HyDrop v2, 10x v2, and sciATAC data. Per cell type, the left plots are generated using the coverage while the right plots are generated using the cut site information extracted from cell-type specific fragment files. e, Drosophila-specific Tn5 bias from Hu et al., (2025) plotted on DARs of neuronal cells, somatic muscle cells, glia, and hemocytes. f, Sequencing efficiency compared between HyDrop v2 and 10x v2 samples. The colors are shown in a, the labels are shown for the top three motifs per cell type. g, Scatterplot of normalized enrichment score of cell-type-specific transcription motifs, dots represent a common motif between sciATAC and 10x v2 data colored per cell type as shown in b. h, Heatmap for top 1000 DARs Yolk cells of 10x v2, HyDrop v2, and sciATAC data. DAR: differentially accessible region.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Generated, Sequencing

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , t -distributed stochastic neighbor embedding (tSNE) of 340,604 cells (18 experiments) HyDrop v2 colored by cell type, no batch correction is applied. Data points were randomly shuffled before plotting. b, VDRC library of transgenic flies with ∼2kb enhancers (pink) upstream of minimal promotor and Gal4 reporter. 2,693 enhancers stage 15-16 evaluated in vivo by Kvon et al. (2014). c, Scanning of ∼2kb enhancers with sliding window (10bp shift) of 500 bp evaluated on cell types seen in a. d, Predicted accessibility of 500 bp windows of VT3067 enhancer from VDRC library. Both, the predictions based on the HyDrop v2 and 10x v2 data are shown with confidence intervals of 10-fold replicate models. e, In situ hybridization of VT3067-Gal4 reporter embryos stage 15 with antisense Gal4 probe by Kvon et al. (2014). f, Prediction (top) versus ground truth (bottom) of 500bp region within 2kb long VDRC enhancer VT3067 active in neuroblasts. g, Nucleotide contribution score of the 500 bp within VT3067 enhancers (VDRC fly line) trained on HyDrop v2 data (top) and 10x v2 data (bottom). Previously described TF binding sites are highlighted for the corresponding TF. h, Nucleotide contribution score of the 500 bp of synthetic enhancer designed to be active in epithelial cells based on sciATAC data by de Almeida et al. (2023): HyDrop v2 and 10x v2 model predictions shown for epithelial cells versus predictions by the original model for somatic muscle cells data. Previously described TF binding sites are marked for the corresponding TF. i, Nucleotide contribution score of the 500 bp of synthetic enhancer designed to be active in muscle cells based on sciATAC data by de Almeida et al. (2023): HyDrop v2 model predictions shown for somatic muscle cells (top) versus predictions by the original model for general muscle cells data (bottom). Previously described TF binding sites are highlighted for the corresponding TF. j, Evaluation of motif importance scores of highly relevant regions for cell type identity indicated by the 10x v2 data-based model and HyDrop v2 data-based model for neurons, glia, neuroblasts, and visceral muscles compared to somatic muscles. An importance threshold of 5 was handled. k, Prediction of tissue accessibility of synthetic enhancer designed for muscle cells in 10-12h embryos. Enhancers were designed by de Almeida et al. (2023) and scored with the HyDrop v2 model. Accessibility is specifically predicted for somatic muscle cells in the HyDrop v2 data. l, Schematic overview of possible footprinting in selected 500 bp region of VDRC library enhancers. Footprinting is performed with seq2PRINT suit by Hu et al. (2025). m, Multiscale footprint of 500bp region in VT3067 in 10x v2 data. n, Multiscale footprint of 500bp region in VT3067 in HyDrop v2 data only. The bottom track shows Tn5 insertion, the predicted Tn5 bias, and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF. VDRC: Vienna Drosophila Resource Center

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Transgenic Assay, In Vivo, In Situ Hybridization, Binding Assay, Muscles, Footprinting, Sequencing

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , Heatmap of correlation of prediction by HyDrop v2 model with ground truth (left), self-correlation of HyDrop v2 data as ground truth (right) for HyDrop v2 data. b , Heatmap of correlation of prediction by 10x v2 model with ground truth (right), self-correlation of 10x v2 data as ground truth (left) for 10x v2 data. c , Model comparison of sequence models trained on HyDrop v2 data (orange) and the 10x v2 data set (blue). The accuracy of the predictions evaluated based on the ground truth of region accessibility is compared on class-wise (top) and region-wise level (bottom) for both the full test set in the data and cell type-specific test sets. d , Predicted accessibility of 500 bp windows of VT3067 enhancer from VDRC library. Both, the predictions based on the HyDrop v2 and 10x v2 data are shown confidence intervals reflect 10-fold cross-validation models. e , Multiscale footprint of 500bp region in VT14978 in midgut cells from HyDrop v2 data only. The bottom track shows Tn5 insertion, the predicted Tn5 bias, and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF. f, In situ hybridization of VT14978-Gal4 reporter embryos stage 15 with antisense Gal4 probe by Kvon et al. (2014). g , Multiscale footprint of 500bp region in VT14978 in midgut cells from 10x v2 data. h , Prediction (top) versus ground truth (bottom) of 500bp region within 2kb long VDRC enhancer VT14978 active in midgut cluster. i , Multiscale footprint of 500bp region in VT40610 in foregut cells from HyDrop v2 data only. The bottom track shows Tn5 insertion, the predicted Tn5 bias, and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF. j, In situ hybridization of VT40610-Gal4 reporter embryos stage 15 with antisense Gal4 probe by Kvon et al. (2014). k , Multiscale footprint of 500bp region in VT40610 in foregut cells from 10x v2 data. l , Prediction (top) versus ground truth (bottom) of 500bp region within 2kb long VDRC enhancer VT40610 active in midgut cluster. m , Multiscale footprint of 500bp region in VT54783 in epidermis cells from HyDrop v2 data only. The bottom track shows Tn5 insertion, the predicted Tn5 bias, and the nucleotide contribution scores based on the HyDrop v2 sequence model. Previously described TF binding sites are highlighted for the corresponding TF. j, In situ hybridization of VT54783-Gal4 reporter embryos stage 15 with antisense Gal4 probe by Kvon et al. (2014). k , Multiscale footprint of 500bp region in VT54783 in epidermis cells from 10x v2 data. l , Prediction (top) versus ground truth (bottom) of 500bp region within 2kb long VDRC enhancer VT54783 active in midgut cluster. VDRC: Vienna Drosophila Resource Center

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: Comparison, Sequencing, Biomarker Discovery, Binding Assay, In Situ Hybridization

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet: a , Prediction of tissue accessibility of synthetic enhancer designed for epidermis cells in 10-12h embryos. Enhancers were designed by de Almeida et al. (2023) and scored with the HyDrop v2 model. Accessibility is specifically predicted for epidermis cells in the HyDrop v2 data. b , Prediction of tissue accessibility of synthetic enhancer designed for gut cells in 10-12h embryos. Accessibility is specifically predicted for midgut cells in the HyDrop v2 data. c , Nucleotide contribution score of the 500 bp of synthetic enhancer designed to be active in gut cells based on sciATAC data by de Almeida et al. (2023): HyDrop v2 and 10x v2 model predictions shown for midgut cells versus predictions by the original model for general gut cells. d, Prediction of tissue accessibility of synthetic enhancer designed for epidermis cells in 10-12h embryos. Accessibility is specifically predicted for glial cells in the HyDrop v2 data (left). Right: Nucleotide contribution score of the 500 bp of synthetic enhancer designed to be active in epidermis cells based on sciATAC data by de Almeida et al. (2023): HyDrop v2 and 10x v2 model predictions shown for glial cells instead showing a repo motif. e , Prediction of tissue accessibility of synthetic enhancer designed for gut cells in 10-12h embryos. Enhancers were designed by de Almeida et al. (2023) but did not give signal in vivo . Here, the synthetic sequences are scored with the HyDrop v2 model showing predicted accessibility in other tissues than the gut.

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: In Vivo

Journal: bioRxiv

Article Title: HyDrop v2: Scalable atlas construction for training sequence-to-function models

doi: 10.1101/2025.04.02.646792

Figure Lengend Snippet:

Article Snippet: As Tn5 insertion inputs, we used the Hydrop v2 and 10x v2 fragment files preprocessed by PUMATAC pipeline for in-house generated data and cellranger-arc (10x Genomics) for public the mouse multiome cortex data .

Techniques: