#16 · Epigenomics
Deep learning methods now annotate chromatin architecture from Hi-C contact matrices
The dominant thread is deep learning applied to Hi-C-derived chromatin architecture. scSTEC annotates chromatin subcompartments directly from individual single-cell Hi-C matrices using two-dimensional embedding and chromosome-wise consensus, without imputation. AxialTAD predicts TAD boundaries from Hi-C matrices with an Axial Contrast Module, reporting 76-80% fewer parameters than deepTAD. HiCP2GAN enhances low-resolution contact maps by using a pretrained Vision Transformer Hi-C foundation model as the discriminator of a plug-and-play GAN. The remaining items, PhenoMe for microscopy image embeddings and ViScore/ViVAE for scRNA-seq embedding quality and multi-scale structure preservation, address adjacent representation-learning problems rather than chromatin structure. What is new here is the shift toward single-cell subcompartment annotation and foundation-model-based enhancement within the Hi-C toolkit.
Published papers
Interpretable models for scRNA-seq data embedding with multi-scale structure preservation.
Introduces ViScore, a scoring framework for embedding quality metrics, and ViVAE, an interpretable deep-learning model for multi-scale structure preservation…
AxialTAD: an axial contrast architecture for TAD boundary prediction from Hi-C.
AxialTAD, a compact deep learning model with an Axial Contrast Module, predicts TAD boundaries from Hi-C matrices with 76-80% fewer parameters than deepTAD at…
HiCP2GAN: a plug and play foundation model-based GAN for Hi-C enhancement.
HiCP2GAN is a plug-and-play GAN that uses a pretrained Vision Transformer Hi-C foundation model as discriminator to enhance low-resolution chromatin contact…
Preprints
scSTEC: Annotating Chromatin Subcompartments from Individual Single-Cell Hi-C Data via Direct Two-Dimensional Embedding and Chromosome-wise Consensus
scSTEC annotates chromatin subcompartments from individual scHi-C matrices via 2D embedding and chromosome-wise consensus, without imputation or cross-cell…
PhenoMe: An interactive AI framework for interpretable, microscopy-driven phenotyping
PhenoMe is an open-source, model-agnostic framework for exploring foundation-model image embeddings of microscopy data, with visualisation, clustering,…
Topic history
| Date | Rank |
|---|---|
| #16 | |
| #15 | |
| #2 | |
| #9 | |
| #32 | |
| #34 | |
| #19 | |
| #3 | |
| #17 | |
| #8 | |
| #2 | |
| #1 | |
| #1 | |
| #2 | |
| #2 | |
| #2 | |
| #12 | |
| #33 | |
| #31 | |
| #12 | |
| #23 | |
| #34 | |
| #9 | |
| #28 | |
| #38 | |
| #15 | |
| #33 | |
| #36 | |
| #24 | |
| #23 | |
| #32 | |
| #29 | |
| #20 | |
| #29 | |
| #38 | |
| #38 | |
| #37 | |
| #36 | |
| #34 | |
| #17 | |
| #21 | |
| #10 | |
| #4 | |
| #6 | |
| #3 | |
| #4 | |
| #1 | |
| #1 | |
| #5 | |
| #32 | |
| #31 | |
| #5 | |
| #6 | |
| #5 | |
| #12 | |
| #7 | |
| #18 | |
| #4 | |
| #19 | |
| #13 |