#2 · Structural bio & protein design
Chromatin structure and nucleosome states inferred from sequencing and contact maps
The dominant thread is computational inference of regulatory state from indirect genomic readouts. CRUSH refines A/B compartments to 1 kb from low-depth data across cell types, single cells, and an ancient mammoth sample. A polymer-physics ensemble and an Activity-by-Multiway-Contact model, built from Hi-C and Micro-C maps, link multiway enhancer-promoter interactions to expression changes in cancer. IDLI classifies long-read single-molecule footprints to detect pervasive nucleosome distortion and TF-linked regulation, while YeIP predicts expression hotspots at yeast integration sites from genomic context. The items do not bridge modalities in the same way; the shared element is mapping one layer of chromatin or sequence information onto another. What is new here is resolution: kilobase-scale compartments from low-depth data and nucleosome-level distortion read from individual chromatin fibres.
Preprints
Similarity-based representation factorization for revealing interpretable dimensions in representational data
Introduces Similarity-Based Representation Factorization (SRF), a method recovering interpretable non-negative low-dimensional embeddings from similarity…
ScaleMAP: Preserving Local Density and Neighborhood Structure in Low-Dimensional Embeddings
ScaleMAP restores local density scale in UMAP/PaCMAP embeddings via change of variables, recovering sparse cell-population bridges.
Faithful Supervised Dimensionality Reduction for Biomedical Data via Decision Geometry
DG-UMAP proposes supervised dimensionality reduction via a low-rank metric deformation from classifier decision geometry, then applies parametric UMAP; tested…
Topic history
| Date | Rank |
|---|---|
| #2 | |
| #4 | |
| #20 | |
| #14 | |
| #11 | |
| #3 | |
| #19 | |
| #1 | |
| #9 | |
| #8 | |
| #6 | |
| #19 | |
| #14 |