Computational Approaches to Analyze Big Root Data Grown in the Field

Alexander Bucksch

Genomes2Fields Workshop @ Phenome 2020

February 24, 2020

1. Food source

2. Sequester atmospheric carbon and provide oxygen

3. Construction material for shelter

4. Energy

Human life relies on plants

PHYSIOLOGICAL

NEEDS

SAFETY &

SECURITY

LOVE &

BELONGING

SELF-ESTEEM

SELF-

ACTUALIZATION

= x

= x

Assumption: If an observed root trait variation is linked

to genes, than the trait is possible to breed

Sa

me

co

mm

on

be

an

ge

no

typ

e

dro

ug

ht

we

ll w

ate

red

Basic idea: Roots change their shape

Phenotype = G + GxE +

E

Compute

on large

data sets

How to link computing, math &

plants?

Unknow

n

phenoty

pic

pattern

Highly

detailed

phenoty

pe

Suxing Liu

Measuring every root in the maize root

system

3D root phenotyping pipeline for field grown maize

Real maize root vs. 3D root model3D root scanner

Recording high-resolution 3D point clouds

8 replicates per genotype

12 genotypes to validate measurements

Traits are measured as averages per genotype

Validation of four traits

Broad sense heritability of all traits

Whole root descriptor distinguishes genotypes in 3D

We can measure a lot – but challenges remain

1. Number of crown root is currently not reliably counted2. Crown root angles are too noisy to be useful3. Number of whorls and distance between whorls sometimes unresolvable

Increase point cloud density throughoptimal positioning of cameras.This means to find an approximationfor the art gallery problem (NP-hard)

Lots of new technology, but what does it help in future?

Limeng Xie

Can we discover unknown phenotypic pattern in big root data?

Limeng Xie

Tagging every

plant

with location

500-1000 times one genotype in a grid

3 genotypes (DOR 364 / L88 57 / SEQ 7)

2 environments water stress / non-limiting

How many root architectures?

What is this mess?

Analyzing D-curves as DS-curves

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Shape curves describe the variation

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Any visible differences between

environments?

Shape curves describe the variation

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

1 genotype + 1 environment = many architectures

An

gle

to x

-ax

in r

adia

ns

Let’s do some math !

How can we group similar curves?

0

1

2

3

4

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Depth

DS_M

ean

Cluster

0

1

2

3

4

5

6

7

8

9

10

11

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Fraction of excavation depth

Result: Number of architectures per data

set

How to group similar curves?

K-needlealgorithm

K-means++clustering

DS-curves From DIRT

Normalized knee plot Optimal k

1 10 20

number of clusters

perc

en

tag

e o

f exp

lain

ed

vari

ation

perc

en

tag

e o

f exp

lain

ed

vari

ation

fraction of clusters

1000 times

What is an outliers for curves?

Modified Epigraph Index(~ variation within the curve)

Mo

dif

ied

Ban

d D

ep

th

(~d

ista

nce

fro

m m

ean

cu

rve)

Method modified from Arribas-Gil, Ana, and Juan Romo.

"Shape outlier detection and visualization for functional data: the

outliergram.” Biostatistics 15.4 (2014): 603-619.

Shape outliers: Curves not following the “obvious” trend of the clusterMagnitude outliers: Curves that peak out of the “typical” bandwidth

DS-

valu

e

Fraction of excavation depth

An

gle

to x

-ax

in r

adia

ns

Fraction of excavation depth

Stability 0.66 0.69 0.70 0.66 0.67

Stability 0.55 0.58 0.50 0.66 0.62

Phenotypic Spectrum of L88 57 (2015)

Water stress

Non-limiting

(n=828)

(n=439)

Stability 0.59 0.59 0.59 0.58 0.60

Stability 0.71 0.70 0.60 0.63 0.74

Phenotypic Spectrum of L88 57 (2016)

Water stress

Non-limiting

(n=304)

(n=328)

Consensus types across years &

environment

Architecture Type 1 Architecture Type 2

Architecture Type 5 (2016)

Architecture Type 4Architecture Type 3

Architecture Type 5 (2015)

Year difference

No difference in shoot biomass (2016/non-limiting)

No significant biomass difference

between architecture types (ANOVA),

Except 2016ww_5-2016ww_3 p<0.05

No difference in shoot biomass (2016/water stress)

No significant biomass difference

between architecture types (ANOVA)

Real time soil water content data in 3D

Volumetric water content regulated with 128 sensors to control 128 sprayers

6 inch deep

15 inch deep

Frac

tio

n o

f vo

lum

etri

c w

ater

co

nte

nt

Frac

tio

n o

f vo

lum

etri

c w

ater

co

nte

nt

Compute

on large

data sets

How to link computing, math &

plants?

Unknow

n

phenoty

pic

pattern

Highly

detailed

phenoty

pe

Collaborators on the presented projects:

Funding for the presented projects:

Acknowledgements

Jonathan Lynch Kathleen Brown Paul Heinemann Dana ChoiJames Burridge

Malcom BennettTony PridmoreSasha Moony

Shawn KaepplerNatalia De Leon

Patompong Saengwilai

Andries TemmeJohn BurkeLisa DonovanJohn Miller

Jonathan Lynch Kathleen Brown Paul Heinemann Dana ChoiJames Burridge

Malcom BennettTony PridmoreSasha Moony

Shawn KaepplerNatalia De Leon

Andries TemmeThursday, 2/275:00 - 5:15 PMPoster 430

Patompong Saengwilai

Collaborators on the presented projects:

Funding for the presented projects:

Acknowledgements

Questions?

Natural History Museum, Mae Rim, TH