ccsws2 9830

Report
Jack Snoeyink & Matt O’Meara
Dept. Computer Science
UNC Chapel Hill
Collaborators
 Brian Kuhlman, UNC Biochem
 Many other members of the RosettaCommons
 Richardson lab, Duke Biochem
Funding
 NIH
 NSF

Scientific Models, esp. for Structural Molecular Biology





Focus on statistical/computational models with


Models are the lens through which we view data
Models are predominantly geometric
Computational models are complex
Models evolve, so testing becomes crucial
a sample source, observable local features, chosen functional form,
fit parameters, & visualization/testing methods
Capture assumptions and date used to build models to:



Visualize for making design decisions while building
Fit parameters to ensure best performance
Record as scientific benchmarks
Case Study: Rosetta protein structure prediction software [B]

Physical and Conceptual models


Kept simple to aid understanding
Statistical and Computational models
Evolve by combining simple models
 Even when complex can still be effective at
Validation (Molprobity) or Prediction (Rosetta)

Spiral development, much like software
 Discover problematic features in some data
 Create an energy function to adjust them
 Fit parameters to improve results
 Check into the software as a new option
 Make default option if everyone likes it
 Occasionally refactor and rewrite, removing
outdated or unused models
But less support for testing…
Our goal:
Capture
data and
assumptions
from model
building for
use in model
visualization
and testing.
Abstraction: A simple component of a complex
computational model consists of:
 One or more sample sources giving


Observable local features having a


Hydrogen bond distances and angles
Chosen functional form that


Pdb files from native or decoys
Energy from distances and angles
Depends on fitting parameters

Weights for combining terms
KMB’03
data set A
data set B
...
gather
features
data set Z
plots
SQL query
ggplot2
spec
filter
transform
statistics
Implemented tools
 Compare distributions from sample sources
 Tufte’s small multiples via ggplot
 Kernel density estimation
 Normalization
Opportunities for
 Statistical analysis
 Dimension reduction …
1400
1200
1000
800
600
400
200
1.
45
1.
55
1.
65
1.
75
1.
85
1.
95
2.
05
2.
15
2.
25
2.
35
2.
45
2.
55
2.
65
2.
75
2.
85
0
[KMB’03]
Histogram of Hbond A-H distances in natives
Scientific unit tests
native, HEAD, ^HEAD
run on continuously testing server
Knowledge-base score term creation
native, release, experimental
turn exploration into living benchmarks
Test design hypotheses
native, protocol, designs
how strange is the this geometry?

Scientific Models, esp. for Structural Molecular Biology





Focus on statistical/computational models with


Models are the lens through which we view data
Models are predominantly geometric
Computational models are complex
Models evolve, so testing becomes crucial
a sample source, observable local features, chosen functional form,
fit parameters, & visualization/testing methods
Capture assumptions and date used to build models to:



Visualize for making design decisions while building
Fit parameters to ensure best performance
Record as scientific benchmarks
Case Study: Rosetta protein structure prediction software [B]

similar documents