Skip to content

Latest commit

 

History

History
283 lines (167 loc) · 4.5 KB

lecture_8.md

File metadata and controls

283 lines (167 loc) · 4.5 KB
title author institute date output
Working with pangenomes
Alexander Leonard
ETH Zürich
Day 3
beamer_presentation
theme keep_tex
Boadilla
true

Recap

We have learned about:

  • determining the core and accessory components of a pangenome
  • assessing the general graph metrics of a pangenome
  • the importance of useful graphs rather than perfect graphs

Objectives

By the end of the lecture, we should be able to:

  • manually edit GFA files
  • understand filtering graphs
  • be aware of emerging pangenome methods

Overview

  • Bandage GFA editing
  • vg toolkit
  • Advanced pangenomics

Pangenome correction

GFAs are just text files.

. . .

We can edit text files however we want.

S s1  AATTTACC
S s2  GGTAT
S s3  T
L s1  + s2  + 0M
L s1  + s3  + 0M
L s2  + s3  - 0M
P ME  s1+,s2+,s3
P YOU s1+,s3+

Pangenome correction

Reference patching, Hi-C contact maps, etc.

Is it hacking or is it curation?

. . .

verkko graph{ width=60% }

vg

Several related commands:

  • vg prune
  • vg mod
  • vg clip
  • vg simplify

. . .

These are widely used, but inconsistently described in methods.

odgi

Can also use odgi to find such regions we want to trust/remove.

odgi depth -W 1000:10:50 -i pangenome/bovine.gfa

The future of pangenomics

The future of pangenomics (as of next month)

. . .

These are mostly pre-printed concepts.

There is little "field" and application literature.

Black box pangenomes

We know pangenomes improve analyses, but some fields are hesitant to change.
So what can we do?

. . .

How much "the algorithm" do we understand for accepted programs?
Some degree of input → black box → output...

Black box pangenomes

vg surject pipeline can:

  • align reads to a pangenome
  • "surject" reads back to a linear reference
  • variant call with DeepVariant from a typical BAM file

. . .

Lose some of the benefit by discarding non-reference alignments, but net positive.

Black box pangenomes

Graph beats linear alignments for GIAB variants.

F1 GIAB{ width=60% }

Black box pangenomes

F1 medical{ width=60% }

Even more so in challenging regions.

Black box pangenomes

Similar to visualisations, linear coordinates can be easier to work with.

. . .

Especially when interacting with collaborators who don't understand pangenomes.

Personalised pangenomes

Imagine a graph containing all variation from a population.

. . .

Currently, too much variation:

  • is prohibitively slow to work with
  • can be detrimental to accuracy (too many combinations)

. . .

How meaningful is rare variation anyway?

Personalised pangenomes

We can create a "personal" pangenome with vg haplotypes.

. . .

Using k-mer information from our sample, we can "pick" the most relevant nodes to keep.

. . .

The resulting diploid pangenome should outperform linear or whole-pangenome alignment.

Personalised pangenomes

Currently, only diploid samples are supported.

. . .

Any approach based on haplotype-paths in the graph is limiting.

. . .

Recombination and many other common biological mechanisms cannot be modelled.

impg

Building bigger and bigger graphs is a huge challenge.

. . .

What if the region we care about is small?

. . .

"Implicit pangenome graph": lift over known coordinates from one assembly to all others.

And then build that pangenome.

impg

We can reuse the wfmash all-versus-all alignment many times.
So the cost is upfront and saved later on.

. . .

Still "reference-free"

impg

Also useful for just looking into assemblies quickly.

S288C#1#chrI        50000  100000
DBVPG6044#1#chrI    35335  85288
Y12#1#chrI          36263  86288
DBVPG6765#1#chrI    36166  86150
YPS128#1#chrI       47080  97062
UWOPS034614#1#chrI  36826  86817
SK1#1#chrI          52740  102721

Bugs bugs bugs

These are powerful new methods, but are not stable.

. . .

You might:

  • be stuck while the developers fix issues
  • have to rerun analyses frequently
  • not even get the code to finish

. . .

Pangenomes are high-risk but (hopefully) high-reward research.

Summary

. . .

:::incremental

  • Expert knowledge can be used to manually fix GFAs
  • Filtering graphs based on depth/rarity is possible
  • Many interesting and promising pangenome approaches are on the way :::

Questions?

And then lunch