Introduction Biological Background Building Models Results Moving Forward
Modeling changes in gene expression in
neurodegeneration in mice
Fourth Annual Primes MIT Conference
Lalita Devadas
Mentor: Angela Yen
May 18, 2014
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
1/18
Introduction Biological Background Building Models Results Moving Forward
Outline
1 Biological Background
2 Building Models
3 Results
4 Moving Forward
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
2/18
Introduction Biological Background Building Models Results Moving Forward
Outline
1 Biological Background
2 Building Models
3 Results
4 Moving Forward
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
3/18
Introduction Biological Background Building Models Results Moving Forward
Gene Expression
Central dogma of molecular biology
Regulated by genetic (ACTG) and epigenetic factors
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
4/18
Introduction Biological Background Building Models Results Moving Forward
Epigenetics
Histone modifications
DNA Enviroment
Epigenetic factors are context that affect gene expression
Histone Modifications
Chemical changes to histone protein core or protruding tail
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
5/18
Introduction Biological Background Building Models Results Moving Forward
Experimental Data
Neurodegeneration in mice
Control Mouse
Neurodegenerative
Mouse
Gene
Expression
Data
Gene
Expression
Data
Change in Gene
Expression Data
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
6/18
Introduction Biological Background Building Models Results Moving Forward
Outline
1 Biological Background
2 Building Models
3 Results
4 Moving Forward
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
7/18
Introduction Biological Background Building Models Results Moving Forward
Types of Models
Random Forest
Random Forest model
Returns value based on set of values determined by a group of
decision trees
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
8/18
Introduction Biological Background Building Models Results Moving Forward
Types of Models
Linear Model
Linear model
Finds a linear correlation between predictors and response
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
9/18
Introduction Biological Background Building Models Results Moving Forward
Two-Step Model
Classification and Regression
Chromatin
Mark Data
(∼3000 genes)
Classification
More Expressed
than control
(∼400 genes)
Similarly Expressed
(∼2200 genes)
Less Expressed
than control
(∼400 genes)
Regression Regression Regression
Predicted
Expression Change
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
10/18
Introduction Biological Background Building Models Results Moving Forward
Outline
1 Biological Background
2 Building Models
3 Results
4 Moving Forward
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
11/18
Introduction Biological Background Building Models Results Moving Forward
Final Model
Train on half of data, test on other half
Classification step: Random Forest model
Regression step: Linear models
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
12/18
Introduction Biological Background Building Models Results Moving Forward
Results of Two-Step Model
Classification Graph
Accuracy of Classification
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
13/18
Introduction Biological Background Building Models Results Moving Forward
Results of Two-Step Model
Classification Values
Expression
Sensitivity Specificity
More than control
.199
.043
Less than control
.063
.043
Same as control
.971
.128
Sensitivity
how good the model is at predicting if a data point belongs in a certain class
Specificity
how good the model is at predicting if a data point doesn’t belong in a certain class
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
14/18
Introduction Biological Background Building Models Results Moving Forward
Results of Two-Step Model
Regression Graph
Accuracy of Regression: r 2 ≈ 0.18
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
15/18
Introduction Biological Background Building Models Results Moving Forward
Outline
1 Biological Background
2 Building Models
3 Results
4 Moving Forward
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
16/18
Introduction Biological Background Building Models Results Moving Forward
Next Steps
Reprocess data (to improve our predictive power)
Use different data (possibly Roadmap data)
Create R package (for cross validation)
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
17/18
Introduction Biological Background Building Models Results Moving Forward
Acknowledgements
I would like to thank:
My mentor, Angela Yen
Prof. Manolis Kellis
Andreas Pfenning
Prof. Li-Huei Tsai and Elizabeth Gjoneska (Experimental
collaborators)
PRIMES program
My family
Lalita Devadas — Modeling changes in gene expression in neurodegeneration in mice
18/18