Journal: bioRxiv
Article Title: Revealing long-range heterogeneous organization of nucleoproteins with N 6 -methyladenine footprinting
doi: 10.1101/2024.12.05.627052
Figure Lengend Snippet: A. Schematic for SMRT CCS and 6mA calling. B. Comparing ipdSummary and ipdTrimming pipelines for 6mA calling using the subread level Sequel data. C. Performance of ipdSummary and ipdTrimming on four different datasets as indicated on left. The thresholds for calling 6mA are indicated by black lines. For dam + plasmid DNA dataset, the thresholds for 98.5% 6mA recall are indicated by red lines. The same recall thresholds are used to calculate the background noise level for 6mA calling in the WGA dataset. False positive rates (FPR, grey shade) and false negative rates (FNR, red shade) for both Tetrahymena gDNA and human 6mA-FP samples were calculated after Gaussian demixing of IPDr distribution. D. Comparing the computational costs, including job all clock time, CPU running time and peak memory usage of ipdSummary and ipdTrimming. E. Comparing 6mA calling results by ipdTrimming and fibertools. Red curves represent IPDr distribution for all adenine sites calculated by ipdTrimming; blue curves represent the number of adenine sites with indicated IPDr values that are called as 6mA by fibertools. The thresholds for 6mA calling by ipdTrimming are indicated by black lines. Recall rates (black label) and FPR (red label) were calculated after Gaussian demixing of IPDr distribution. F. Comparing 6mA calling results for the Revio system, generated by the Sequel kmer model, human native gDNA control, and the Revio kmer model. FPR (grey shade) and FNR (red shade) were calculated after Gaussian demixing of IPDr distribution.
Article Snippet: The CCS IPD value was subsequently compared to a reference IPD value of its unmodified counterpart embedded in the same local sequence; all reference IPD values were generated by a pretrained kmer model ( https://github.com/PacificBiosciences/kineticsTools/tree/master/kineticsTools/resources ).
Techniques: Plasmid Preparation, Generated, Control
![Overview of the Attention-Guided Erasing (AGE) Methodology. a Self-Supervised Pretraining using <t>DINO</t> [11]: A teacher student ViT framework, leveraging a teacher-student ViT-S self-distillation framework. b AGE [13]: Attention head visualizations from the SSL <t>pretrained</t> teacher ViT-S are converted into binary masks to isolate key ROIs and then used to erase background regions. c Transfer Learning with AGE: AGE is used on the input images using each of the attention heads with a random probability during training. The attention head yielding the highest validation performance is selected for final AGE-based transfer learning](https://pub-med-central-images-cdn.bioz.com/pub_med_central_ids_ending_with_9719/pmc11929719/pmc11929719__11548_2024_3317_Fig1_HTML.jpg)
