#BoG18: Talk Notes

32 min read bog18, bioinformatics

Intro

Very sparse and poorly written notes covering #BoG18.

Typos everywhere. Things may change dramatically over time as I scan back through notes.

I’ve tried to respect #notwitter. Will be updated periodically.

Speaker (Last Author)

Genome Engineering and Genome Editing (Tuesday Night)

Jef Boeke

Writing Genomes

Building synthetic yeast genomes. Contig/chr one by one. All designed. Sc2.0

80+% complete for each of the 16.

“dark matter”

Can we use ‘big dna’ to functionally query mammalian genomes? ‘Synthetic haplotypes’

Building dif combinations of haplotype blocks

synthetic hypervariation:

- query enhancers
- alt splicing

built 102kb locus (human) and put into yeast

- built in 3kb chunks and can assemble dif combiations

big dna

Can build big dna pieces CEGS grant will build 3 100kb+ loci / year want community input

Greg Findlay (Jay Shendure)

Accurate classficiation of thousands of BRCA1 variants with saturation genome editing

vous a problem BRCA1:

  • 4243 clinvar snvs

  • 50% VOUS

How to functionally validate?

Use Homology-direct repair (HDR). Can engineer precise edits.

Use a library of SNVs for the HDR

Over time selecdtion removes non functional edits

Each experimente:

  • millions of cells
  • millions of sequencing reads to count SNVs

Variable effects at splicing junctions

  • sometimes just 2-3 bp
  • sometimes 9 base pairs

also have matched rna-seq data

aberrant splicing causes RNA depletion

matches up really well with clinvar designations

question:

  • hdr effeciency rate? 10-90% effectiveness

Stephen Levene (Andrew Fire)

eccDNA is a possible mediator of chromosomal polymorphism at multple loci

‘physical chemist by training’

  1. coli genome: 1 femtoliter volume

100 fold compaction problem (DNA)

~10k fold for mammalian

Techniques for DNA/chromiatin flexibility:

  • Hi-C
  • FISH
  • SLICE (Beagrie et al Nature 2017)

eccDNA (circular dna outside chromosome/nucleus(?))

  • unclear how it forms
  • elevated levels associatd with genome instability

how to capture?

  • DNA -> SDS lysis -> isolate gDNA -> cscl gradient -> bottom bit
  • or exoV treatment (leaves circular alone)

take pictures of the loops with high resolution microscopy

sequenced a few (unfortunately with short read illumina)

modeled with molecular dynamics

David Truong (Jef Boeke)

Resurrection of Histone H3k27 Me in brewer’s yeast by human prc2 and plant atxr6

human pathway reconstruction in yeast

  • avoid pleiotropy (hopefully)
  • more real than in vitro

Yeast (s. cerevisiae) lost histone mods

Adding them back???

yes

humanize yeast histones

add synthetic human histones

force out wt histones with +5FOA

20 days later ….. one colony

Keep growing the colony out

WGS: mutations in cell cycle regulation

  • bypassing histone cycle checks?

brewer’s yeast does not have H3K27 methylation

  • PRC2 complex (to methylate h3k27)

add the stuffs - what happens?

  • made artifical chr with PRC2 complex
    • and a slightly broken one
  • not much happens in WT yeast
    • no me changes
  • deleted H3k36me3 (might antagonize artificial chr)
    • nope
  • can you jump start with atxr6 (does mono me)?
    • yes (confirmed with mass spec)
      • not super high levels (0.054% tri me)

Feng Zhang

Advances in genome editing technologies

two major classes of CRISPR

  • class 1 (multi subunut)
  • class 2 (single subunit crRNA-effector)

trying to find new class 2

  • bioinformatic screen with BLAST of cas1
  • found a bunch (Shmakov collaboration)

cas13

  • added into e. coli
  • modify to only edit RNA?

rna editing

  • reversible
  • nuclease based editing inefficient in post mitotic cells
  • dCas13 linked with ADAR (adenosine to inosine) + guideRNA –> A to I conversion in RNA
  • 90+% conversion
  • 1732 off target incidents
    • 925 off target with non-targeting guide!
    • so protein itself is not ideal….
    • identified non-binding residues of ADAR
      • mutated them
  • v2 works better
    • 18385 off target (v1) to 20 (v2)
  • still developing

Molly Gasperini (Jay Shendure)

my fav of the night

crisprQTL mapping as a genome-wide association framework for cellular genetic screens

lots of guideRNA to made mutations, check for dif in expression

nuclease inactive cas9

want to test all enhancers against all genes

scRNA-seq + guideRNA (multiplex gRNA)

  • thus multiple perturbations per ‘assay’
  • 15-30 / cell!

targeted 1,119 candidate enhancers

  • 15 guides / cell
  • 47k cells
  • 10X
  • CROP-seq
  • works really well
  • crisprQTL usually targets closest gene
    • sometimes not….
  • matches up with histone chip-seq
  • 34.3kb average distance from enhancer <-> gene

** new data! ** 4,801 enhancers

  • built logistic regression model on pilot to pick new candidates
  • 30 guides / cell
  • correlates with pilot

manolis kellis q:

  • why doesn’t work so well? expected more
  • what about multiple SNPS / block?

Eilon Sharon (Hunter Fraser)

Testing genetic var effect on fitness using precise genome editing

high throughput edigint

crispey: cas9 retron precise parallel editing via homology

use bacterial reverse transcriptase and RNA retron to covalently link ssDNA donor to guide-tracrRNA

can insert long sequences

(yeast)

measure fitness of genetic variants (growth competition)

sequence every 2-3 gens

model at linear relative strain abundance / time (generation #)

# missense var ~ # synonymous var for effecting fitness!

Luca Pinello

CRISPR-SURF exploratory and interactive software for analyzing CRISPR-base tiling screens

Uncover non-coding functional regions

** Nice overview of CRISPR tiling strategy ** Mutate (tile across region) -> Measure pheno change (somehow) -> Assess (sequence gRNA)

No unified framework to analyze these kind of assays

many challenges

  • biological noise
  • sgRNA efficiencies
  • non-uniform spacing
  • perturbation / assay differences
  • epigenetic perturbation can be wide (changing 200bp or so)

deconvolve with generalized lasso

fastq -> score -> segmentation -> deconvolution -> region ID

Population Genomics (Wednesday morning)

Mattias Joakobsson

** out of my field here **

Sequence based approaches utilizing complete modern and ancient genomes to investigate early human history

Use full genomes on ancient pops

Population divergence models

                      X
     +                X
     |                X
     |                X
     |                X
     |               XXXX
     |              XX  XXX
time |           XXX      XX
     |         XXX         XX
     |        XX            XX
     |
     |      A B              C
     |
     v

Can model whether discordant or concordant (a,b,c) over time

estimate pop divergence (time) in generations

use genes from different populations to estimate divergence

  • ‘tt method’

stone age humans from sourthern africa

  • 13 genomes from 3 people
  • a bit ‘right’ of yoruba
  • admixture with east africa missed something

Jaemin Kim (Elaine Ostrander)

Genetic Selection of Athletic Success in Sport Hunting Dogs

WGS of sport hunting (10 breeds), terrier (i breeds), and ‘village’ dogs (unselected - an outgroup)

  • 14 million SNPs

59 genes under strong selectdion in hunting dogs (compare to terrier and village)

  • blood circulation GO terms
  • and a bunch of ‘process’ GO terms

ASIC3 - resistance to muscle fatigue?

  • maybe?
  • a guess based on known gene function (I think)

dogs do agility performance competitions

  • made a metric to find breeds good at winning
  • WGS of 92 breeds of 299 dogs
  • ROBO1 significant 3e-4 (FDR corrected? Don’t know)
    • neuronal migration, axon guidance
  • 1243 SNP chip
    • dogs classified by agility performance
    • ** do only pure breeds do agility? **
    • ROBO1 SNP AF increases with more winning breeds

racing speeds (whippet)

  • not ROBO1
  • TRPM3 (1.6e-3)

CDH23 - increased tolerance to loud noise and low startle reflex - do hunting dogs have poor hearing?

Useful stuff maybe for competitive dog breeders

Q (Kellis?): polymorphic nature of traits across dogs. what’s the question?

A: complex traits, incomplete answers right now

Q (Kellis?): enrollment bias for dogs that will win

A: tried to control by grouping breeds

Q: what kind of mutations?

A: mostly noncoding (answer in LD I guess)

Q: project personality onto dog … look at dog behavior/traits relating to this?

A: try to objectively test dogs (can’t trust owners….)

Elaine: people developing stanrdard tests for dogs (yes, owners lie)

Ipsita Agarwal (Molly Przeworski)

Widespread differences in the mtation spectrum of X and autosomes

Males contrigute more germline mutations than females

  • epigenetic differences for gamete development (methylation)
  • sperm in mitosis all the time
+
|
|
|                         XX
|                       XXX
|                    XX
|                  X
|               XX
|             X
|       X XX
|   XXXX
|  XX
|                 XXXX
|   XXXX XXX  XXXXX
|
+------------------------+

Mutation rates get wider as males age (top line male, bottom female)

  • eyeball 3x worse?

GnomAD:

  • 120 million SNPs
  • 60% singleton (50%) and doubletons (10%) for variants

Looked for X-autosome difs

X/A diversity = (mutations(x) / all X) / (mutation(a) / all a) (a is autosome, x is chrX)

Bootstrap test for mutation types to make null distribution

  • big shift in X (more than expected)
  • T->A and C->A more common in X versus autosome

Replication timing

  • inactive X has more mutations

Enriched of C>G (meiotic recombination / DSB) mutations

Amnon Koren

Genetic architecture of human DNA replication origin activity

We have extensive maps of human genomic / epigenomic

But where are replication origins?

  • yeah….good q
  • yeast have them
    • yeast have a DnaA/OriC signatures
  • how to find?
    • many techniques - don’t agree well (or at all)

different parts of genomic replicate at different rates

  • can measure coverage across time, right
    • yes
  • sort G / S phase
    • check coverage
  • Did in 2012 with human - but not precise enough (low resolution)

Is this a polymorphic trait?

  • skipped cell sorting
  • works well enough
    • and way faster to do
  • uh, wait, this has been done already (if cells are growing)
    • again, yes, LCL from 1000G

But still, did WGS >140 hESC lines

  • oooooo, reproduces REALLY well
  • find ‘master’ ORI that are pretty much always present
    • crucial regions in replication?

GWAS of DNA replication timing

  • ‘rtQTL’ (replication timing)
  • big hit on chr7
  • 756 with FDR < 0.1
  • most fall within replication origin
    • direct relationship, then, cool
  • getting causal SNPs with CAVIAR
  • enriched with active chromatin states

perhaps some QTL stabilize TF binding motifs - stuff happening in motifs

Q: look for associations with structural variation? (1000G data)

A: looking at this now

Q: cell type specific? (thank you)

A: 20-30% are dif across cell types (cool)

Q: is ORI piggybacking on enhancer / regulatory system? (or other way around??)

A: maybe (or other way around)

Q: cis effects - did you find any trans or pleiotropy?

A: nothing strong (spurious stuff maybe?)

Q: HiC/3C data profiles comparisons?

A: not yet

Sarah Tishkoff

Novel loci associated with skin pigmentation identified in African populations

Integrative omics of copmlex traits

  • epigenomics
  • transcriptomes
  • microbiome
  • metabolomics
  • proteomics
  • genomics

LOTS OF POSTERS

200 WGS Aricans - 35 million SNPS - 20% novel

81% GWAS european

Skin color is adaptive trait - spectrophotometry of skin color - and take DNA - boom GWAS - 1600 people - found 8 regions

SLC24A5

MFSD12 - novel - transmembrane transporters - found enhnacer activity - functional work! - KO mRNA in melanocyptes - get more melanin - colocalizes with lysosome - ZF KO - yellow gone - mouse KO - diff colors - looks like gr/gr mouse - 9bp deletion in MFSD12!

DDB1 - DNA repair in UV damage - pigmentation in tomatoes - fine mapping hits TMEM128 - luciferase assay with enhancer activity - huge haplotype blocks of low het in europeans/asians - selective sweep, near complete fixation

OCA2/HERC2 - exon10 SNP alt splicing - rs1800404

convergent evolution of very dark skin - african and south asia

q: speculate about lysosome?

a: pheomelanin made in lysosome (like) structure?

q: surprised to not find var close to mcr1 a: no

q: chimp with no hair has vitiligo?

a: dont’ think so - have been assured chimps have light skin

Patrick Albers (Gil McVean)

Non-parametric estimation of allele age for variants in pop-scale seq data

Want to know history of allele at single locus

Genealogical approach (GEVA)

  • look for coalescent events
    • concordant and discordance allele pairs
  • use HMM to detecdt haplotypes segments
    • non parametric
  • assess allele age…not sure how
    • incorporate time some how?
    • model with real data to pick parameters??

model - simulation with simple pop, const size, fixed mut rate - compare estimate with actual - good cor (0.953)

also tested in error-filled data - haplotypes, errors, phasing issues - still works, but overestimate young alleles (think they are older)

ran in simons genetci diversity project and 1000G - oldest var ~37.5k years in AFR - SAS 12.5K - EAS 11K - AMR 7.5K - EUR 6K (why so young? - admixture)

cumulative coalescent decoding (CCD) - what frac of your genome share with another genome back in time? - ran this pair-wise across 1000G dataset

will release a genome-wide atlas of allele age >16 million variants

Laura Hayward (Guy Sella)

Polygenic adaptation in response to sudden change in environment

What time scale? Mentioned Human expansion out of Africa, which is ~100k years

stabilizing selection reduces pheontypic variance (omega is width of trait distribution) - kingsolver et al. 2001

model change of phenotpye in environment change

OK…not following this talk well at all (pretty sure it’s me). Laura is copiously using cartoons, which is usually works for fools like me.

not cetain whether this stuff is driven by theory or data or both - no, no data - equations from first princples

conclusions: - polygenic adaption is rapid - short term: large effects drive change - long term: moderate effect alleles replace them

topol Q: is this like dogs? big changes in short term

a: no, not modeling dramatic changes like this

Functional Genetics and Epigenomics

Job Dekker

Folding, unfolding, and refolding genomes

How does the genome work? structured

Dekker et al. 2002: 3C. I remember this paper. And totally failing at doing this tech myself

A/B compartments….TAD…..enhancer - gene loops. One slide summarzing many publications and of cool work.

                                TAD
                                                loop (cohesin-mediated)
                 loop (cohesin-mediated)       XXXX 
                XXXX                         XX   XX
              XXX   XXX                     XX      XXX
            XX        XX                  XX           XX
          XX            XX              XXX              XX
        XX                X           XXX                  XXX
      XX                   XX        XX                      XX
    XXX                      XX    XXX                         XXX
  XX                          XXXXX                              XX
 XXXX                           X                XXXXXXXX       XXXX
    XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX       XXX X XX
ctcf                           ctcf                                 ctcf

Interphase / metaphase - interphase has structure - metaphase erases structure (fat diagonal of 3c map)

meiotic chr fold as helical nested loop arrays (helix with loops coming off) - Gibcus…Dekker Science 2018

Using ATAC-seq, cut&ruhn to assay CTCF binding patterning in interphase / metaphase

FRAP to measure stable CTCF binding in interphase / metaphase - unstable in metaphase

Q: CTCF sites occupied in mitosis - why? Nucleosomes taking over?

A: Good q, don’t know. Think nucleosomes ‘sliding in’

Q: missed it

A: protein levels of CTCF doesn’t change during what phase cell is in

Q: what is keeping the promoters open during mitosis

A: don’t think pol is bound….promoters more open to begin with….something bookmarking the site????

Flora Vaccarino

Integrative multi-omics analyses of iPSC-derived brain organioids

#notwitter

Carninci

RADICL-seq: novel tech for genome-wide mapping of RNA-chromatin interactions

Many many (~20k) functiona lncRNA

But what is role? Activate genes, promoter, enhancer? Repression of genes? Establish insulation?

RADICL-seq - capture RNA-DNA interactions with crosslinking (formaldehyde, 1-2%)

Where do they map?

  • Lots in the intron
  • open chromatin
  • 380k RNA-DNA interactions
  • enriched in TF family members
  • looks like lots of trans interactions
    • but way fewer than cis,as you would expect
  • some genes like MALAT1 interact with entire genome
  • compartments sort of like TADS
    • weak cor with Hi-C (0.27)

Johnathan Griffiths (Berthold Gottgens)

Charting the Diversification of Mammalian Cells at whole genome scale

Gastrulation focus (mouse)

350 whole embryos

Collected every 6 hours: e6.5 to e8.5

10X chromium, 94k cells (post QC), 15,000 UMI (median), 3.5k detected genes (median)

Big t-SNE plot

  • talk of ‘direction’ and ‘trajectory’ which I dislike for t-SNE…
  • but can back up with time point data

Clustered with association of cell types to each other

Chimera embryos

  • KO gene, but only chimerically (inject KO mESC into blastocyst)
  • whoah
  • can compare KO cells vs not KO cells within full atlas
  • damn

Emma Farley

Regulatory principles governing enhancer specificity during development

Otxa (neural enhancer)

How do enhancers encode function?

Need to test in embryos across time

Ciona

  • have notochord, heart

enhnacer + prom + gfp -> electroporation == inside embryo

made 2.5 million synthetic enhancers (barcoded)

electroporate 100k eggs -> mRNA -> sequence (remember, we have barcodes) -> identify functional enhancers

  • pooling of embryos?
  • doesn’t seem like it…but seems like you have too for $

Can we we make inert enhancers functional with small tweaks (change to optimal seq)

  • lights up EVERYTHING
  • need mix of optimal and sub-optimal sites to maintain proper expression

spacing between enhancers also important

  • adding just a few bp between motifs >>> expression

interplay between spacing and motif ‘strength’ (canonical-ness)

and orientation (flipping motifs can break function)

suboptimization as design parameter hilarious and terrifying (great, let’s make things worse to control things more precisely)

wut - SHH enhancer that causes polydactyly….points out that SNP makes binding site better

Deboever et al. Cell Stem Cell 2017 D’Antonion et al. Nature communication 2017

Jake Yeung (Felix Naef)

Clock dependent chromatin topology modulates circadian transcdription and behavior

promoter - enhancer loops

cry1 24 cycle in liver (high at night, low at day)

how are enhancers used?

4c-seq on mice collected every 4 hours for 24 hours

  • contacts change over time

h3k27ac changes also at same location

removed that region in mouse: cry1deltaE mouse

  • clock runs faster (15 mins over 24 hours)
  • corresponding mRNA difs in cry1

Minal Caliskan (Casey Brown)

Genetic and epigenetic fine mapping of complex trait associated loci in the human liver

Genotype <-> RNA-seq <-> H3K4me3 <-> H3K27ac 50 <-> 50 <-> 10 <-> 10 (people)

Found eQTLs specific to liver (vs GTEX)

Found hQTLS also (histone)

eQTL + hQTL + RNA-seq to ‘fine map’ GWAS loci

  • blood pressure
  • coronary artery disease

Parisa Razaz (Talkwoski)

Tissue-specific molecular sigmnature of 16p11.2 reciprocal genomic disorder

engineer rgd with CRISPR (make microdels and microdups)

iPSCs models and mouse models with tx profiling (brain and not brain)

Evolutionary and Non-human genomics

Monica Justice

Informing human genetic variation and therapeutic entruy poitns through modififer screens in mice

Rett syndrome

  • x-linked
  • 1/10,000 live female births
  • developmental regression
  • fatal in males

MeCP2

  • ubiquitously expressed (highest in neuron)
  • regulate gene expression
  • chromatin modifier

can a second mutation improve pathological phenotype?

  • forward genetic unbiased screen in mice
  • how to assess phenotype???
  • random mutagenesis (ENU)
    • of males
  • evaluate symptons to make a ‘health score’
    • eek, so a LOT of work
  • screen 3200 mice

exome seq G1 founder male for candidate phenotype - 20-99 new lesions - backcross - mate 10 G2 - analyze 10 G3 offspring

many pathways affected

  • including DNA repair and lipid metabolism

sum3 line (null allele of Sqle)

  • dysregulation of cholesterol pathway
  • brain lipd turnover fails
  • other lipids overproduced

Other mutations:

  • NCoR1/SMRT
  • HDAC3

Interventions:

  • Statin drugs to decrease cholesterol biosynthesis
  • clinical trial just finished

Sum6 line: Rbbp8 (aka Ctp1, Sae2)

  • DSB (double strand break) repair
    • MeCP2 null alleles prone to DNA damage…
  • BRCA1 as cofactor
    • found mutations in this too that surpress symptons

combination therapy

  • DSB + lipid metabolism
  • with mouse mut can ‘cure’ disease

great talk

q: radio sensitivity?

a: yes, higher DSB. cortex highly effected, even in cells that don’t divide

Arang Rhie (Erich Jarvis, Adam Phillippy)

Genome10K Vertebrate genomes Project

generate error-free reference-quality gnomes assemblies for all 66k extant vertebrate species

why? because bad ref genomes are super confusing to work with when YFG (your favorite gene) is missing

  • bad assembly a major contributor

goals

  • 1 Mb N50

  • QV40

  • both haplotypes (eek)
  • big scaffold

  • 90% reads to right chr/contig

only 16 genomes approaching these goals (NCBI, exclude human/mouse)

name drop awesome goat paper from phillippy group

  • PacBio + Hi-C + Illumina + Optical Map

Phase 1:

  • 200 orders, 266 species

Process:

  • PacBio for contigs
  • Scaffolding with: 10X, Bionano, Hi-C
  • Gap filling with PacBio (and base polishing?)

Evaluation:

  • gEVAL (Chow et al. 2016)
    • k-mer com-leteness
    • synteny
    • gene completeness (BUSCO Simao et al 2015)
    • KAT
    • Mashmap2
    • Evol.Highway

Data Release!!!!!

Heterozygosity causes allelic dup

  • zebrafinch very high
  • using trio binning for phasing

q: annotations?

a: we have a working group….

Olga Dudchenko (Erez Lieberman Aiden)

DE NOVO assembly of mammalian genomes with chr-length scaffolds from short reads for <$1000

pointing out that unnamed company 1k genome requires a good reference

  • this is de novo

tile reads…contigging…but of course doesn’t work too well when you hit repeats

can order contigs with linking data (scaffolding)

  • EXPENSIVE and slow

Hi-C

  • 3d genome sequencing
  • make genome-wide contact map
  • has been used to reassemble known genomes
    • often with MANY other techniques

using hi-c contact map to correct contigs/scaffolds

  • showing different types of issues (translocations/inversions/etc)
  • GUI work to correct assembly?!
  • they have something more automated?
    • seems like would be nuts to do with just short reads (100k + pieces)

Open to collaborations (http://www.aidenlab.org)

100 Million HiC PE150 illumina + 300M DNA-seq PE150 (w2rap)

3D-DNA suite

Kasper Munch

Multiple selective sweeps removed Neandertha admixture of the X chr in out of Africa pop

strong selective sweeps in great ape in chr X

because of inter chromosomal conflict between x and y?

165 male X chromosome from Simons Genome Diversity Project

  • SGD gertting name dropped a ton this conference

large proportion of very similar haplotyes among non-african people

  • big left spike on density plot
  • don’t see in african pop

what are these haplotypes?

  • low divergence haplotype defined as dist less than 5e-5 of >25% people in 500kb sliding window
  • locations grouped with ethnicity in chrX

find 500kb sweep spanning non-africans

  • after leaving africa
  • but before world-wide spread

does it overlap human-chimp common ancestor?

  • yes

swept chr regions have less archaic admixture - Skov et al 2018 bioRxiv for method

trying to get time window of sweep

  • 45k years

why?

  • don’t know (for certain)
  • ‘crazy ideas’
    • two pop leave africa
      • only one meet neanderthal
      • then merge
      • many small groups leave
      • later one big one leaves, which meets neanderthals
      • then all merge

Gavin Sherlock

Joint distribution of fitness effects for beneficial mutations in yeast

(did a bang up job of keeping first half of session on time)

Interested in adaptive (positive) mutations that improve fitness

yeast

‘experimental evolution’

Lineage tracing system:

  • random primers
  • plasmid library
  • barcodes
  • generations (with replicates)
  • sequence barcodes after each serial transfer
    • beneficial mutation barcodes will increase
    • most adapative lineages are rare (<<1%)

genotype to fitness map (ID beneficial mutations)

‘narrow in scope’

  • conditions not very diverse
  • what are tradeoffs?

‘achieving high fitness across a wide range of habitats is apparently hard’

  • mutations deleterious
  • direct tradeoffs (antognistic pleiotropy)
  • lower fitness elsewhere

what’s the balance of these three effects?

So…

  • measure fitness in dif environments
  • m x n matrix of fitness in environment x alt environment
  • exponential scaling….

added second barcode to mark condition

evolve populations -> isolate colones -> POOL (second barcode!) -> measure pools across n environments

  • pleiotropy common
  • big cluster with variety of performance in alt conditions
  • a few generally useful clones
  • many generally bad

no discussion of variants / genes that increase fitness?? Boooo

q: look at structural variation?

a: whole chr dup very common to increase fitness for a few gene. hard to get structural data with short reads

q: super yeast generalists? what about cycling yeast across conditions repeatedly? genetic basis of super yeast?

a: have done a bit of flipping across environment. diploids more fit. really dancing around the genetic basis question

Update

Bit unfair to Gavin here (I was just really hoping for some speculation on gene function!). His tweet response:

Anne Ruxandra Carvunis

The genome’s reservoir of benefical proto-genes

mechanism of molecular innovation

proto-genes

  • non-genic seq <-> pervasive expression -> adaptive potential -> novel gene

yeast genome pervasively tx and has lots of ‘random’ ORFs (open reading frames)

RiboSeq (Ingolia et al 2009) to ID translated seq

  • 2k ‘proto-genes’
  • match up with RNA-seq
  • and predicted ORF

artifically evolve by systematic overexpression (of proto-genes)

  • and put WT yeast and check to see who wins over time
  • so measuring growth

find all three classes:

  • beneficial (some)
  • deleterious
  • neutral (most)

enrichement of proto-genes in beneficial class (OR ~3 pretty good)

found 6 generalists

‘network integration’

  • what are good proto-genes predicted function
  • #notweet
    • (not 100% certain in results yet)
    • happy that Anne is sharing a bit about general proto-gene functions
      • compelling stuff
  • bit confused on network integration …. didn’t see any evidence or discussion of it
    • maybe a syntax issue?

Elaine Ostrander

Catalog of 722 Whole Genome Sequences Reveals Variants Controlling Morphology in Domestic Dogs

dog GWAS (with WGS)

  • easy to do GWAS with low n (do high n lineages)
  • easy to find LD block
  • hard to find mutation

Parker et al. Cell Reports 2017

  • breed relatedness with neighbor joining tree (up to >200 dog breeds)
  • found haplotype sharing across breeds

91 million variants

  • 60% low VEP impact
  • argh exploded 3d pie charts!!!

Lots of filtering, pop stratification ….

  • 14 million vars

GWAS on the 14 million

  • 28 associations
  • big table, little text
  • validate on known biology (Furnishing, fur length, height)
    • yes they find the stuff they expect

LCORL

  • strong influence on dog size

SMAD2, HMGA2, IGF1, IRS5, IGSF1 (bigger size/weight)

  • also for longevity (anti correlated)

ESR1 (leg length)

lincRNA (near MSR83) (drop ears)

gorab, chsy3 (tail shape)

Working to get to 10K dog genomes (Dog10K)

Bobbie Cansdale (Claire Wade)

3D modeling of Hi-C data to investigate spatial organization of the canine genome

TADs - modular units of genome struture/organization

Regions assessed: - MITF - ASIP - RALY - TYRP1 - MC1R - KRT71 - RSPO2 - MSRB3 - IFG1

Qualitative look at genes known to do stuff…and what genes are near them (in Hi-C TAD block)….as far as I could tell.

3D modeling….haven’t done yet….

Cancer and Medical Genomics

Trey Ideker

Decoding patient genomes through the hierarchical pathway architecture of the cancer cell

‘somatic’ eQTL RNA expression + WGS + enhancers (mapped to genes with GeneHancer) + covariates to calc eQTL Found ~190 (including positive control TERT)

yeast now

ppi (protein protein interactions) - Dutkowski et al 2013 nature methods

DCell - neural network guided by hierarchical cell bio - 3500 cell subsystems (from ppi?) - 12 layers - 12 million yeast genotypes (single and pairwise KO) - http://d-cell.ucsd.edu

DCell a model for what to hope to do one day for human

Rajbir Batra (Carlos Caldas)

Decoding the dynamics of DNA methylation in breast cancer

Cluster of breast cancer with:

  • gene expression
  • CNV
  • miRNA

Little epigenetics

So looking at Me

METABRIC data source

  • 1482 primary breast tumor
  • 237 matched adjacent norm
  • RRBS (reduced representation bisulfite seq)

Background drift of methylation

  • tried to pick background regions (not promoters)
  • late replication timing associated with higher me

Now look at more functional region

  • but control for background drift

Change mean methylation from tumor to matched tissue

  • ‘epiallelic burden’
  • heterogeneous across tumor types

Patrick Short (Matthew Hurles)

Contribution of de novo mutations in the regulatory elements in neurodevelopment disorders and autisum

DDD Study (Deciphering Developmental Disorders)

  • 8k trios
  • exome

Model germline mutation rate

  • Samocha et al. 2014
  • enrichment in missense and PTV

What proportion of remaining cases may be regulatory?

Regulatory elements

  • heart enhancers
  • VISTA
  • Conserved elements

Look for variation in elements

enrichment in fetal brain conserved elements

manhattan plot not so good (flat)

  • variety of reasons
    • enhancers shorter
    • less consequential than proteins
    • poor understanding of enhancer code

can we development noncoding constraint score for deep whole genomes?

25,000 whole genomes

mutation rate vary across genome / chr / region

using random forest to model and use a large number of features

  • not so successful (see below)

~160 mb of open chromatin is contrained, most of this is conserved nucleotides in poorly conserved elements

thinking that 1 million deep whole genomes to accuratly model small (tfbs) sized elements

selection on nucleotide level, not ‘peak’/chunk level

Max Shen (David Gifford)

Predictable and precise template-free editing of pathogneic mutations by CHRSPR-Cas9 nuclease

High-throughput assay (1872 guideRNA) and deep sequencing to assess mutation pattern

1262 genotypes / target site

  • mostly deletions (almost 90%)
  • lots of ‘microhomology’ deletions (60%)
    • deletion with multiple equal quality alignments

built dnns to predict mutation pattern

  • three models for microhomology, not microhomology, and insertion (knn for this one)
  • inDelphi
  • 90% outcomes predicted (70% single bp resolution)

so mutation patterns are semi-predictable (semi is my language)

use model to design guideRNA to get more preditable outcome

nice web app - can’t find it online…not ready yet - **https://github.com/maxwshen/indelphi-dataprocessinganalysis**

Sharon Plon (PJ Lupo)

Cancer risk among children with non-chromosomal birth defects in the genetic ovrerlap between congenital anomalies and cancer in kids (GOBACK) study

1/33 children born with birth defects

GOBACK

  • look for overlap between cancer and genetic abnormalities (birth defect) in children

10 million births 0.5 million birth defects 15k cancer 1.8k birth defect and cancer - napkin math would expect 750

more birth defects = greater chance of having cancer

different defects have diff cancer increase rates

can go back to the children

  • doing WGS

Massa Shoura (Andrew Fire)

eccDNA-mediated ‘scars’ in the TTN gene may contribute to microfiber diversity

Coding regions ‘shedding’ the eccDNA ‘circles’

Including Titin

Looked at cardiomyocyte vs iPS vs lymphocytes to see if eccDNA shedding changes

  • unique to cardiomyocytes

eccDNA shedding seems dynamic in different cardiomyocyte conditions

Sidi Chen

Towards mapping functional cancer genome atlases

Concept:

  • cancer seq data to make cancer atlas
  • make mouse models

Marcin Imielinski

Signature of complex structural variation across thousands of cancer whole genomes

Diversity in rearrangments across cancers

Successful at finding SNV signatures

  • Alexandrov et al 2013 Nature

How to count rearrangments????

  • lots of types (dup / insertion / translocation /etc)

gGnome (R packages):

Computational Genomics (! - I’m very excited)

Oliver Stegle

Methods for the joint analysis of high-dimensional traits and samples substructure in human cohorts

project high dimensional data into lower dimensions

  • confounding factors
  • inference of sparse factors (f-scLCVM)
  • spatio-temporal dependencies (SpatialDE)

https://github.com/bioFAM/MOFA

  • scheme for merging data

Identification of sub-gropu specific genetic effects

  • interation between environment/behavior (self reported??) and genotypes

random effect model to match genetic structure to environment

scaling up - with big 2D matrices

StructLMM: mixed model interaction

  • phenotye = G + GxE + additivity(?) + noise
    • Casale et al PLOS Genetics 2017

BMI on UK biobank

  • 240k unrelated people
  • 7.5 million variants
  • 64 ‘environments’
    • not certain what this is

Increased power for detecting GxE (loci match GWAS ID’ed loci)

How to deconvoluate which environmental var drives GxE?

  • include/exclude environmental vars to rank change in bayes factor
    • reminds me of variable important in random forest

“Fresh results”

  • can leverage GWAS info with GxE analysis
  • genome wide search
  • GxE of 7 metabolic traits
    • FTO…MC4R….SEC16B
    • loads of hits (nice manhattan peaks)
  • correlation structure
    • loci x loci spearman cor
    • cluster blocks by loci
    • by traits not so much
  • % variance explained
    • adding up to 10% var

https://www.github.com/limix/struct-lmm

q: pleiotropy? metabolic traits confounding

a: yes (we see overlap between traits)

q: environment matrix. how sensitive to variable weighting?

a: controvery over how to weight. quite robust for a not crazy amount of traits

Ben Strober (Alexis Battle)

Modeling genetic effects during cellular differentiation

Genetic effects dependent on cellular environment (or cell type)

  • tx studies are generally single time point

study dynamic genetic effects during cell diff in iPSC

  • ID varying genetics effects
  • new stats approaches

Cell Diff time course

  • iPSC (14 lines)
  • 16 time steps (ranging 0 to 15) when diff to cardiomyocyte
  • == 217 total RNA-seq samples

Marker genes going up (or down) over time as iPSC –> cardiomyocyte

  • time is pc1
    • percent of var?

cisEQTL in each time step

  • WASP combined haplotype test (Geijn Nat Meth 2014)
  • 50-200 eQTL genes (FDR <0.1) per step
  • replicates to independent iPSC eQTL study (Banovich Gen Res 2018)

Temporal dynamics - correlation of eQTL pairs summary stats - hey look - blocks

sparse matrix factorization to find shared eQTL effect size (across time) patterns

this compression is correlating to time step blocks

id dynamic-QTL

  • GLM to predict allele-specific read counts
    • log(y nj) == intercept + time + genotype + (time x Genotype)

rs1897133

  • association gets stronger over time
  • nice integrated model instead of doing tons of independent tests
    • I assume the GLM gets much higher power

obligatory slide showing enrichment in cis-regulatory elements (chromHMM enhancer)

‘is time the best way to quantify differentiation progress?’

  • no
  • cell lines will develop at dif rates
  • hmm to infer (predict) ‘differentiation progress’
    • hmm states (4 states over 15 days) work

future

  • more lines
  • more assays (single cell, ATAC)
  • apply methods to other longitudinal data

Meena Subramaniam (Jimmie Ye and Noah Zaitlen)

Population-scale single cell seq to reveal context specific effects in lupus

(bulk) gene expression var could be due to diff cell type proportion or state

  • answer: scRNA-seq

SLE (lupus)

  • PBMCs (convenient)
    • Meena did justify this cell type

10X chromium:

  • 1 well / donor $1300
  • 200 donors / 200k is … expensive
  • 50k cells with 25 people / well
    • $200 / donor
  • pooling
    • double rate increase
    • lose sample ID

multiplexed droplet scRNA

  • cell barcode
  • then can demultiplex (with genetic var)

genetic barcode

  • unique variants in sample
    • but scRNA is mega lossy
    • prob have to use many variants?
      • how many?
        • 20 gives 98% ID

doublet ID (with ‘demuxlet’)

  • bayesian model to find doublets
  • intentionally merged samples
    • can deconvolute robustly

mux-seq and dumuxlet

https://www.github.com/statgen/demuxlet

https://www.satijalab.org/costpercell

did human immune ‘census’ (is this the new atlas?)

  • 4k cells / donor (750k total)
  • 8 major cell types t-SNE looks good
  • sle vs healthy have unique locations (not clean blocks / ‘clsuters’)

CD14+ and CD4+ proportions change

  • EHR confirms with CBC (cor 0.83)
  • cool to get to check out the EHR data

novel monocyte population

  • ‘intermediate monocyte’
  • found interferon correlation

stimulate cells with interferon

  • large shift in cell t-SNE patterning

change in both proportion and state

search for genetic factors (GWAS)

  • flat manhattan plot

cis-eQTLS

  • much better power (QQ plot)
  • hundreds eQTLS found per cell class
  • cool to see cell-type specific eQTL

data sharing (June)

q: mech for change in variance

A: rare cell state….or dif enhancers being used with dif noise profile

q: are you looking for novel gene signatures for cell groups?

a: following up on that now

Ricardo D’Oliveira Albanus (Steve Parker!!! <– Hey I know this guy)

Information theory of ATAC-seq data predicts local chromatic kinetics and reveals novel aspects of gene regulation and genomic org

http://theparkerlab.org

ATAC-seq –> open chromatin –> footprints

  • but not all TF leave footprints
    • don’t bind long enough
  • CTCF, AP-1 bind a long time
  • GR short time
  • know from FRAP assays
    • but super hard to do

can we use ATAC-seq data to infer kinetics of TF-DNA interactions?

  • can we break dependence on footprints to predict TF-DNA binding?
    • OMG, can we???

model theoretical binding of nucleosome / TF

  • ‘V plot’
    • CTCF has a nice strong one
    • NFKB less strong

measure chromatin information content

  • f-VICE (feature V-plot Information Content Enhancment)
  • calc f-VICE for each TF
    • AP-1 and CTCF has highest f-VICE score

BMO uses ATAC-seq signal and co-occurring motifs for TF binding prediction

  • compare to CENTIPEDE, DNase2TF, HINT, bedtools intersect
  • f-VICE not involved???
    • feel like a shift happened quickly in the talk
  • BMO works better with (assessed with F1 of predictions)
    • works better for bad binders…?
    • compare /cor with f-VICE
    • so yes, BMO not using f-VICE (I think)
  • CENTIPEDE next best (pretty close)

f-VICE has 8.5 change in comparing CTCF (- cohesin) and CTCF (+ cohesin). - ~10-20X diff by FRAP

No github?

Christina Leslie

Decoding immune cell dysfunction

In solid tumors

Does epigenetic state of tumor-speicif T cells present a barrier to immunotherapy

  • Want to know whether checkpoint blockade approach will work or not in a patient
  • It does in mice (Philip Nature 2017)
    • during development see epigenetic state change (plastic to fixed as time goes on)

intro over, new stuff

Is CD8 T cell tolerance to self-antigens epigenetically encoded? Can tolerance be broken?

model of CD* t cell tolerance to self-antigen

  • TCR(GAG) xAlb:GAG (mouse) –> TCR(GAG) x Alb(GAG) give self-tolerant CD8 T cells
  • two states
  • native t cell (memory) and naive t cell (tolerance)

ATAC and RNA-seq

  • naive clusters from efector/memory/tolerant
  • enriched in putative enhancers
    • is this ever NOT the case??
  • 1837 peaks differntial accessible (out of 95k)
  • showing browser plots with disappearing signal

correlation between accessiblity and gene expression

  • very cool plot
    • RNA log2FC on y axis, genes on x axis (each individual point colored by ‘tolerant’ or ‘memory’)

can you break tolerance to become functional?

  • transiently by transfer hosts (Schietinger et al Science 2012)
  • massive expression diff when you transient rescue
    • but no epigenetic changes
    • audible whoah
  • but how can you really rescue?
    • if you immunize in black 6 background
      • now epigenetic state changes

GLM to find epigenetic signatue of tolerance state

  • peak accesbility log2FC == A + P + T
  • Lef1 (very cool because not DE by RNA)

Pejman Mohammadi (Tuuli Lappalainen)

Using rare allelic expression data for studying rare disease biology @pejminister

Interpreting non-coding genome

  • regulatory var important in disease/phenotypes
  • big place, poorly annotated (or not at all)
  • no clear measure of regulatory constraint
  • in comparison coding regions
    • codon triplets
    • pLI, etc.
  • want to something similar, but for noncoding

regulatory variation and expression outliers

  • rna-seq noisey
  • maybe ASE (allele specific expression)
    • haplotype specific
  • try to find outliers
    • (plot ref allele abundance / alt allele abundance for each gene)
    • not linked to populatin info!

how to set alleleic imbalance cutoff?

  • check GTEx
  • but ASE patterns….are diverse (reg effect, allele freq, LD, ….)
  1. define quant measure of reg var
  2. ref. population: estimate regulator var for each gene
  3. check in patient

log alleleic Fold Change (aFC)

  • Mohammadi et at Gen Res 2017

plotting different outcomes

  • aFC vs experssion FC
  • copy loss, wildtype, reg variant,

“I’m skipping 20 of my favorite slides”

How to model when so much info is missing?

  • binomial model
  • ANEVA (Analysis of Expression Variance) in GTEx data
  • tissue specific estimates

D(g) / D(t) measure regulatory tolerance (**g and t are subscripts - I can’t latex quickly*)

  • ID Haplotinsufficient genes
  • better than noncoding RVIS
  • good heritability on independent data
  • better than noncoding metrics (ROC)

Use skeletal muscle GTEx with Macarthur muscle disoorder cohort (~70)

  • find some D(g) outliers
  • found 20 ‘dosage outliers’
    • similar to GTEx (totally dif datasets)
    • 6x fewer hits than naive binomial test

Erik Garrison (Richard Durbin)

Variation graphs for efficient unbiased pangenoimc seq interpretation

want to represent ‘reference’ genome as a graph

  • to include variation
  • simulate genomes by walking along paths
  • retain info with ‘metadata’ for each point

another model is have every base seen as a big vector

  • certainly cleaner and simpler to visualize

‘pan genomics’

applications

  • 1000g pan-genome graph
  • build and index 1000g graph
    • “computation is tractable”
  • performance of pan-genome worse than paired end seq against linear ref
    • but better when a read has a variant
    • no bias
  • Erik is being explicit about limitations and strengths; very nice

yeast var graph

  • with long reads
  • graph does better overall than linear genome (5% better in mean identity)

ancient dna

  • helps out a lot

graphical pangenomics

  • back to the linear visualization mentioned earlier
  • I really like this
    • I’m very freaked out by spaghetti plots of graphs

Masa Roller (Paul Flicek)

Tissue specific enhancer and promoter evolution in mammals

enhancers <-> promoters <-> gene expression

using chip-seq as proxies for enhancer / promoters

create comprehensive regulatory maps

  • 4 tissues
  • 10 mammals
  • 3 histones
  • 3 replicates
  • 360 ChIP-Seq!
    • 4 dropouts
    • require min seq depth
      • 20 million good reads for H3K27ac, H3k$me3
      • 40 million for H3k4me1

promoter == H3K27ac and H3K4me3 enhancer == H3k27ac and H3k4me1

look between tissues/species

  • UpSet plot!
  • testes has lots of tissue specific promoters
  • but still many shared promtoers

tissue specific regulatory elements evolve more rapidly than tissue shared elements

  • promoters and enhancers show this patterns

regulatory elements rarely switch activity between tissues

  • what do switches look like?
  • lower H3K4me3 and lower H3k27ac
    • maybe weak promoters?

see switches across evolution

  • again, not too many
  • some promoter <-> enhancer (10-20% by my eyeball)
conference bog18 bioinformatics talks

Related

comments powered by Disqus