Download presentation
Presentation is loading. Please wait.
1
Integrative Analysis of Biological Data Sai Moturu
3
MAGIC M ultisource A ssociation of G enes by I ntegration of C lusters Goal: Integrate heterogeneous types of high-throughput data for accurate gene function prediction Bayesian reasoning Incorporates expert knowledge Yeast Data
4
Integrative analysis ! Why ?? High throughput methods sacrifice specificity for scale High throughput methods sacrifice specificity for scale Microarray data alone is good for hypothesis generation but lacks specificity for accurate gene function prediction Microarray data alone is good for hypothesis generation but lacks specificity for accurate gene function prediction By using heterogeneous functional data, the prediction accuracy is improved By using heterogeneous functional data, the prediction accuracy is improved
5
Need for MAGIC Studies have combined different types of data in a heuristic fashion on a case by case basis Studies have combined different types of data in a heuristic fashion on a case by case basis No general scheme or probabilistic representation is applied No general scheme or probabilistic representation is applied Methods for combination of specific data Methods for combination of specific data MAGIC – general method to integrate disparate data sources MAGIC – general method to integrate disparate data sources
6
Input to MAGIC Input: Gene-Gene relation matrices for each data source Input: Gene-Gene relation matrices for each data source The elements of the matrix are scores that indicate whether there could be relationship between two genes The elements of the matrix are scores that indicate whether there could be relationship between two genes The score can be binary, discrete or continuous The score can be binary, discrete or continuous Input format is flexible and allows genes to be in more than one group or cluster Input format is flexible and allows genes to be in more than one group or cluster Thus does not exclude biclustering or fuzzy clustering methods Thus does not exclude biclustering or fuzzy clustering methods
7
Structure of the MAGIC Bayesian network Prior probabilities assessed by experts Prior probabilities assessed by experts
8
Evaluation No gold standard for gene groupings exists No gold standard for gene groupings exists GO is the best available reflection of current biological knowledge GO is the best available reflection of current biological knowledge Use a cutoff of 3 levels in the hierarchical structure to say that to genes are functionally related Use a cutoff of 3 levels in the hierarchical structure to say that to genes are functionally related
9
Results
10
Results
12
AVID A nnotation V ia I ntegration of D ata Integrates data to build high-confidence networks in which proteins are connected if they are likely to share a common annotation Integrates data to build high-confidence networks in which proteins are connected if they are likely to share a common annotation AVID predictions functional annotation in all three GO categories AVID predictions functional annotation in all three GO categories
13
AVID stages
14
AVID results
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.