the mapping problem : how do experimental biological models relate to each other, and how can...

The Mapping Problem: How do experimental biological

models relate to each other, and how can dynamic

computational models be used to link them?

Gary An, MDUniversity of Chicago

2014 MSM Consortium SatelliteSept 5, 2014

Bethesda, MD

A Tale of Two Mappings

Seok, et al.

Genomic responses in mouse models poorly mimic human inflammatory diseases.

Proc Natl Acad Sci U S A. 2013 Feb 26;110(9):3507-12.

Takao and MiyakawaGenomic responses in mouse models greatly mimic human inflammatory diseases.Proc Natl Acad Sci U S A. 2014 Aug 4. pii: 201401965.

The Multi-scale Translational Challenge

Organism

Organs

Tissues

Cells

Molecules

Genes

Barriers to Understanding

Multiscale/MultiStep

Investigation

Biological Experimental Workflow

In vitro

In vivo

Clinical

Where “ “ represents inferred knowledge

Q1: What is the justification for this translation of knowledge?Q2: What is “similar?”

Basic Set Concepts(*From Wikipedia)

Biological Paradigm (Legacy of Zoology)

“Animals are classified as follows:

1. those that belong to the Emperor, 2. embalmed ones, 3. those that are trained, 4. suckling pigs, 5. mermaids, 6. fabulous ones, 7. stray dogs, 8. those included in the present classification, 9. those that tremble as if they were mad, 10. innumerable ones, 11. those drawn with a very fine camelhair brush, 12. others, 13. those that have just broken a flower vase, 14. those that from a long way off look like flies.

Celestial Emporium of Benevolent Knowledge – Jorge Luis Borges' fictional taxonomy of animals from his 1942 short story The Analytical Language of John Wilkins.

Bio Paradigm 1: Sets of Components (Detailed

Descriptive)

In vitro

In vivo

Clinical

Components

Components

? ?

Partial Functions at best because Bio Models are

Opaque

Bio Paradigm 2: Sets of Functions/Behaviors

In vitro In vivo Clinica

lInjective Injective

? ?Mapping More Conserved/Preserved*

Q: What is the “nature” of the injection? => Identify the Description of the Behavior/Function*Note: This does not mean components are not included, but rather are abstract representational components

The Role of Modeling Part 1

In silico PS1

Bio Mode

l

ExplicitInjection

• This Injective function is Explicitly Described (Model specification/structure)

• As a Dynamic Model, the In Silico Model represents behavior

The Role of Modeling Part 2

In vitro

In vivo

Clinical

Injective Injective

? ?

In Silico PS1

In Silico PS2

In Silico PS3

Bijective Bijective

Explic

itIn

ject

ion

Explic

itIn

ject

ion

Explic

itIn

ject

ion

The (Potential) Danger of Modeling

In silico #2

Bio Mode

lBijective

• Higher Fidelity directed of Bio Proxy Models move towards Bijective Relationship (Output vs. Generative)

• Risk: Too closely approximates non-mapping Bio Proxy Models => Can’t Serve as Bridge

The Failure of Modeling

In vitro

In vivo

Clinical

Injective Injective

? ?In

Silico #1

In Silico #2

In Silico #3

Bije

ctiv

e

Bije

ctiv

e

Bije

ctiv

e

? ?

The Translational Goal of Modeling

In vitro

In vivo

Clinical

Injective Injective

In Silico PS(n)

Q: Can the In Silico Model now be used to explore behavior space not reachable by In vitro/In vivo Models?A: I claim “Yes”

Modeling for Personalized Medicine

Patient 1

Patient 2

Patient 3

Injective Injective

In Silico PS(n)

Embrace Heterogeneity!

Take Home PointsBehavior/Function Maps highly conserved

Dynamic In Silico Models serve as bridgesConserved In Silico Model Structure => Encapsulate Transferable Knowledge/Hypothesis of MechanismOutput Heterogeneity => Different ParametersComponent Mapping now related to role in Behavior Generation

In Silico Models for Bridging cannot be made too precise => over tuned/fit/mapped

Explains/Utilizes Biological Heterogeneity

Pathway to Translation and Personalization