Structure Only ROC Curve

1
(No Transcript)
2
(No Transcript)
3
(No Transcript)
4
Structure Only ROC Curve
5
StructureOtherInfo
6
(No Transcript)
7
Approach

• Model 874 proteins related to human diseases
• Examine an additional 369 disease related
  proteins with experimentally determined
  structures
• Map all known SNPs onto the structures

8
Results

• As compared to average residues and nsSNPs,
  disease related mutations are more likely to be
  found in the protein core
• Disease related mutations tend to be clustered
• These results held up in both the modeled and the
  experimentally determined structures

9
(No Transcript)
10
(No Transcript)
11
Disease mutations affect protein-protein
interactions?
12
Application of support vector machines to predict
DNA-binding sites in DNA-binding proteins Igor
B. Kuznetsov Zhenkun Gou Run Li Seungwoo
Hwang GenNYsis Center for Excellence in Cancer
Genomics, University of Albany
13
SVM Construction

• Used Ahmad and Sarais dataset of 62 DNA-binding
  proteins
• Inputs included amino acid sequence, evolutionary
  information (PSSM), structural information
  including secondary structure, solvent
  accessibility, and spatial neighbors

14
Results

• 78.9 accuracy
• 0.6 correlation coefficient (if I recall
  correctly)

15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
Kernels for time series data

• Exploit the fact that measurement i follows
  measurement i 1
• Model the time series data as a dynamical system
  data points are now trajectories
• Kernel function is the dot product of each pair
  of trajectories

19
Results

• Apply classifier to problem of predicting a
  patients response to a drug (Multiple Sclerosis
  patients taking human interferon beta)
• Dataset expression profiles of 70 genes from 52
  patients at up to 7 time points over 2 years

20
(No Transcript)