SCGD 2024

all the talks are with preprints

Genome wide TF binding

SHARE-Seq

In the aim of interpreting the footprint, in my words, to build a DNA footprint database –> trans-element <– DL (inspired by dl)

How stable are the footprint as DNA are stable?
can it output pleipotrip footprints?
the scanning params (padding)

–> use case identify TFs that drive cell aging –> nucleosomes shift

3D spatial from “raw” 2nd seq

–> 3D reconstruction based on more raw photonic data –> expand the in situ seq by introducing the expanded hydrogel –> data on microns or DNA distance from lamin

coupling with lamin to study the chromatin structure is a convention

inactive “X” as an exampled biology to be studied with this

Sumamry of chromatin resolution and techs

Discussion

ATAC replace ChIP

but it’s based on knowledge from ChIP

open-ST

20,000 genes

from Nikolaus Rajewsky lab

“3D virtual tissue block”
module cell segment, module transcripts, map modules

biological output

3D boundary between cell types, for example, tumor cells and other cells

Discussion

• sequencing-based has more varied complexed noises. Ideal for hypothesis generating
• imaging-based is ideal for validating the hypothesis.

Rare cell type by na cytometry based PERFF-seq

Coleb Lareau

study strategy

how much they are differ from other non-rare type?

Criteria for sucessful assay

Based on

1. RNA FISH
2. 10x Genomics Flex scRNA-seq, where cells are fixed

   I didn’t understand how seq on fixed cells bu two probes on one read

benchmark using FC/RNA cytometry enrichment

Biological output

1. enrich cells with TF expression
2. enrich with coding and ncRNA
3. rare cell types enrichment
4. FFEP and frozen tissues

Discussion

1. easier with RNA vs DNA
2. initiate probes? ~ size of target gene
3. how many markers? limited by cell numbers but not markers

I would say is the choice of orthogonal markers

Single cell perturbation dictionaries

1. Chemical (cytokine stimulation)
   • Cui, wt al, Nature (2024)

     I read this quite carefully already

2. Perturb-Seq
   • first cytokine stimulation
   • then fix and permeablize for perturb-seq (3 guides per gene for 40-60 genes)

Analyzing

1. binarize perturbatoin output as in Nat Genet (2021)
2. imporve to continous score

spatial and sc by Saturn and star-map

github
paper

convention

1. smoothing I missed how
2. sc analysis neighbors are also spatial neighbors

solution

1. sample from multiple neighborhoods
2. smoothing
3. sc analysis

spatial-resolved sc genomics of human brain by Xiaowei Zhuang

1. space of cells in tissues
2. space of intra-cellular
3. space of chromasome

Techs

1. MERFISH > 1k genes 2019 Technology

   1. error-robust bc
   2. combinatorial label
   3. sequential imaging

2. MERFISH extened to 3D genomics 2020

localize DNA loci > 1k genomic loci
+ > 1k nascent RNA
+ protein

1. extended to epigenomics

with antibody to Tn5 taged protein
in a gel so with proximal info
–> epigenetic markers

1. thick tissue 3D merfish

200 $\mu m$

biology

1. Mouse brain cell atlas. related pub: Zhang et al., PMID 38092912 (2023) somatic contact, paracrine signaling or short-range interactions compose cell-cell interaction
2. Human brain cell atlas

Discussion

human mouse diff (behaviral)

• not only from dense (human < 3x mouse)
• modulize effects
• loci to genomic to get chromosome organization by imaging
  • techniqually no problem to increase
  • high density in nucleus as a rate limiting –> more runs, so tedious
  • worth efforts
  • now mega base resolution –> higher resolution, you need more runs and sts, imaging time

I didn’t think of a way to get around this density

Fixed Universal cell embeddings by Yanay Rosen

Concepts

• e.g. put cells in one universal waddington landscape
• AI e.g. from CZI

with this, we could model cellular change in silico

Represent cells with - representable expression matrix, e.g. low-dimensional embeddings - a sentence then can study with NLP - feature ordering? - a bag of RNA (no order)

I would consider this is prior-free - transformer Represent genes - Excel (redicule) - Gene2Vec - chatgpt

Solution:
Gene represent adopting an aa model ESM

Fixed:
NOT finetune, so called 0-shot **Self-supervised

therefore

Output can be massive, and provides a universal space to be mapped

how high dimension is this bag of RNA representation space

TIME modality by ZMAN-seq from Ido Amit

basically time stamped

Makom & Zman-seq

benchmark spatial seq techs

Methods of the Year 2020

Framework to identify Good data?

1. targets not overlap across methods map to sc and benchmark sc
2. spatial more abundant than sc, why
   1. lack of spcificity: barnyard-like plots to assess specificity
   2. lack of spcificity: coexpression of distinct cell type markers (mutually exclusive)
   3. confound with sensitivity: counts per cell correlate with segmentation size
3. sensitivity and specificity bu controlling for segmentation differences

   how?

Qang, Huang, et al.,

Xenium, Merscope, CosMx from FFEP

1. sensitivity: by spatial vs sc
2. specificity: %genes above background [x] Xenium

Cook, et al.,

Spatial autocorrelation as a measure of quality. hypothesis: spatial patterning

1. probe detection
2. spatial patterning
3. correlation with sc
4. marker

open questions

1. tardeoff panel size vs quality
2. segmentation
3. generality

   if this is necessary?

Discussion

1. segmentation problem, sts, molecules inaccuratelly assign is one reason, autofluorescence is another among many others
2. recommend barnyard in mouse tissues (staining info from DAPI) is a good approach

Universal recording of cellular interactions by Sandra Nakandakari-Higa in Mucida Lab

LIPSTIC based on antigen on immune cells

Universal LIPSTIC

this is good for study syncytiation

[x] quantitative measurement of interaction [x] outcome of interaction?

Disccusions

stability, label from 1-3 h depending on tissue and cell type

My impression

CHANLLENGES make this field worthy devoting

• Previous
  Mother or Child?
• Next
  Algernon: outside of the boundary of human science and ethics