Making Medicine more personal, precise, and predictive

Is so-called clinical BigData actually sufficient?

Alexander Hoffmann

http://www.signalingsystems.ucla.edu

http://www.qcb.ucla.edu

Medicine is about people

… and molecules !

Human Health is multi-scale complexity

DNA & genes

chromosome epigenetics

molecular networks

cellular function

cell-cell interactions

organ function

Human Health is multi-scale complexity

Billion-fold length scales

Human Health is multi-scale complexity

Billion-fold time scales

100

The Groundhog

Feb 2 • cloudy

• sunny

March • early Spring

• prolonged winter

Biomolecular measurement

Disease outcome Whether drug works

Statistical Model

Predicting Human Health

RNA

statistical model

Bio

marke

rs

Diagn

osis/Th

erap

y

Some biomarkers are very accurate

Sickle cell anemia: 100% accurate

Some biomarkers are quite accurate

BRCA1/2 mutation

And we understand how they work

But predictive power of many biomarkers is limited - diseases are complex -

Salles et al 2011

Combining multiple measurements can improve predictive power

Lou Staudt NCI

Combining multiple measurements can improve predictive power

Lou Staudt NCI

Combining multiple measurements can improve predictive power

genome

clinical records

statistical model

molecular data / biomarkers

improve diagnosis

personalize therapy

reduce costs

clinical p

rese

ntatio

n

Big Data in Biosciences

The Precision Medicine Opportunity: Predictive power based on correlations derived from previous patients

genome

epigenomes transcriptomes

proteomes metabolomes

improve diagnosis

identify patient cohorts for studies

personalize therapy

reduce costs

improve treatment/ drug pipeline

clinical p

rese

ntatio

n

microbiomes

clinical records / UC BRAID

statistical model

The Holy Grail:

Precision Health Care

genomes

epigenomes transcriptomes

proteomes

clinical records

metabolomes

statistical

model

prevent disease

life style adjustment

statistical

model

food intake exercise

sleep patterns environmental

microbiomes happiness/stress

mHealth monitors heart, pressure, temperature, sugar, perspiration, urine

prophylactic treatment

precision treatment

earliest intervention

The Holy Grail: Precision Health Care What will it take?

The predictive power of statistical models is a function of

the amount and quality of training data.

The Holy Grail: Precision Health Care What will it take?

How much training data ?

Genome Environment

Consider the variation

Life Style

10 Million SNPs (variable locations)

210,000,000 = 103,000,000

Millions of variants

Millions of variants

106 106

>>>> people ever on the planet

x x = “gazzilian”

Epigenome

Is there an alternative approach?

Airplanes: amazing complex but predictable

Boeing 737: 367,000 parts

Humans: 20,000 genes, 200,000 proteins, 37 × 10^12 cells

1 fatality in 2 x 10^9 miles flown: safest mode of transport

Airplanes are so predictable that flight can be simulated !

The reason for airplanes predictability:

Based on physical principles, and these are described in mathematical equations

“Bernoulli equations”

Two Algorithmic Worlds

Statistical Mechanistic

a + b = c

Agnostic of - Knowledge - Mechanism of causation - Scale

P(a,b) –> c

Precise, but requires knowledge

Predicting Human Health

RNA

statistical model

Predicting Human Health

RNA mechanistic

model

mechanistic

model

mechanistic

model

mechanistic

model

Identify components: parts list Learn physical principles that connect them Learn how each part works in context of the system Build computational model and simulate Physiological functions

Developing Mechanistic Models

2015: NIH budget: $30 billion 300,000 researchers 2,500 institutions

0

5.000.000

10.000.000

15.000.000

20.000.000

25.000.000

30.000.000

35.000.000

1950 1960 1970 1980 1990 2000 2010

The NIH has invested > $500,000,000,000 into biomedical knowledge

DNA & genes

chromosome epigenetics

molecular networks

cellular function

cell-cell interactions

organ function

Biology / Life Sciences: at every scale

Molecular Biology

Biochemistry Cell

Biology Physiology

Biomedical knowledge

a

b e

IB IB

AAAAA

IKK

IB IB

IB

IKK

IKK

IB

Inflammatory Signals

NFB target genes

= k13 * NFBnucleus

+ k-4 * IBaNFBn

+ k-5 * IBbNFBn

+ k-6 * IBeNFBn

+ k7 * IBaNFBn

+ k8 * IBbNFBn

+ k9 * IBeNFBn

- k-13 * NFBcytoplasm

- k4 *

IBa NFBn

- k5 *

IBb NFBn

- k6 *

IBe NFBn

d NFBn

dt = gain - loss

Nuclear Import

Dissociation

Dissociation

Dissociation

Bound IB Deg.

Bound IB Deg.

Bound IB Deg.

Nuclear Export

Association

Association

Association

The challenge of extracting knowledge

x x

x

x

x

x

x x

x x

x

x x x

x

x

x

x

t

h

Are knowledge-based models necessary?

R2=acceptable t √h

1. Knowledge-based models allow for predictions beyond the training range.

Newton h

t

x

2. Knowledge-based models interpret confusingly complex data

Where is the outlier?

Y

x x x x x x x

t

Y

Y X

Knowledge: the system oscillates

Are knowledge-based models necessary?

The person who will become very sick.

??

t

Y

3. Knowledge-based models allow for predictions when relationships are non-linear. - Essential for Drug Dosing

Y X

Are knowledge-based models necessary?

Precision Health Care

Genomes

Behavioral factors

Environmental factors

Molecular diagnostics Epigenomes

Algorithmic Synthesis

KNOWLEDGE

big data

mHealth Monitors

Life Sciences

Medical School Engineering

Health Systems

38

Awesome measurement

capabilities

Electronic clinical data

Computational Modeling

Predictive Biosciences that accounts for the dynamic,

heterogenous genotype-phenotype relationship at all

scales, e.g. cells, organs, organisms, ecosystems.

“Precision medicine” that combines

“personalized medicine” and “predictive pharmacology”,

culminating in “virtual clinical trials”.

The promises of

Big Data Big Knowledge based on

Molecular Biology Biochemistry

Genetics

Research Training - Data generators need QC skills: workshops - Masters and Professional Masters Programs

Education - Graduate Program in QC Biosciences = Bioinformatics, Biomedical

Informatics, Computational & Systems Biology - All Biosciences Graduate programs require more quantitative courses - Flagship Undergraduate Programs in Bioinformatics and Computational

& Systems Biology will help catalyze changes in Life Sciences BSc’s 39

Research - Excellence in Bioinformatics & Genomics - Excellence in Dynamical and Stochastic Processes - Knowledge-based modeling for big-data analysis across scales - Catalyze and enable large grant applications - Collaborate to provide computational expertise - Spin out technology, methodology, tools, biomarkers

Predicting Immune Responses

RNA

mechanistic

model

mechanistic

model

Predicting cell populations based on molecular network

Marcelo Behar

Max Shokhirev

signalingsystems.ucla.edu

http://www.qcb.ucla.edu