The Functional Influence Network (FINE) is a framework for studying gene networks from multiple perturbations data. In difference from existing approaches, which aim to reveal the network of interactions between genes, this method was developed for the deduction of functional networks, describing how given cellular functions are carried out by a set of genes. In this network description, the genesí state determines a quantitative phenotype of the network and its architecture visualizes and explains how the genes, acting together in functional pathways, actually carry out the studied function.
The FINE algorithm was developed to produce such a description of a system, given a set of multi perturbation experiments (for further description of such experiments, see MPA summary). This algorithm utilizes a fundamental result from the field of game theory according to which every function with a discrete domain can be decomposed to a sum of the marginal contributions of all possible subsets of elements in the domain(Grabisch et al., 2000). To this end, we express the studied phenotypic function as the sum over the marginal contributions related to each subset of genes (acting together as functional pathways). Next, we focus only on the most influencing subsets of genes (viewed as functional pathways), ending up with a compact, accurate and intelligible representation of the system studied. We refer to this representation as the CFN (Compact Functional Network). Taking one step backwards, it can now be said that the goal of the FINE algorithm is in fact to reconstruct the CFN.
Several studies were made, exploring the abilities and limitation of the FINE approach. The following list depicts some examples:
- Simulated data analysis - we have performed a
comprehensive set of experiments using simulated multi-knockout
performance data, testing the FINE capabilities in various scenarios. In
our study we have focused on the following questions :
- Having performed a set of different multi-knockout
experiments, how well can we expect to understand and describe the
underlying functional structure of the system?
- Specifically, how does the performance of the
algorithm depend on the predictive accuracy (corresponding to predicting
the missing multi perturbation experiments) of the data that has been
- How and to what extent does this relation depend on
the complexity of the studied system?
- Genetic multi-knock out data analysis - applying the
FINE algorithm to multi-knockout experiments of the DNA Post-Replication Repair
(PRR) system of the yeast Saccharomyces cerevisiae .
- Simulations based on biological data - questioning the
relation between the accuracy of the FINE and the level of predictive accuracy in
biological systems (as in the computer simulations). To this end we use both
the PRR multi knockout data and the putative model of the sea urchinís Endo16 gene
cis-regulatory logic (Yhu et al., 2001). Our results show (at least for the biological systems we have
considered) that the relations between predictive and descriptive accuracy in
biological data are quite similar in nature to those what we have found via our
computer simulations .
A. Kaufman, A. Keinan, I. Meilijson, M. Kupiec, E. Ruppin, Quantitative Analysis of Genetic and Neuronal Multi-perturbation Experiments, PLoS Computational Biology, To Appear
A. Kaufman, M. Kupiec and E. Ruppin, Multi-Knockout Genetic Network Analysis: The Rad6 Example, Computational Systems Bioinformatics, CSB2004
N. Yosef, A. Kaufman, E. Ruppin, Inferring Functional Pathways from Multi-Perturbation Data, submitted
We have created a Matlab(R) FINE package, including: |
The package is freely available for academic use. To acquire it, please e-mail Nir Yosef.
- An Implementation of the FINE algorithm designed to serve as a plug-in module for an application of multi-perturbation data analysis.
- A standalone simulator which can be utilized for making the first steps of working with the FINE package.