Quantitative assessment of biological impact using overlap methods

Abstract

Scores for particular network models (those having a source node connected to a set of measurable downstream nodes via causal edges) are computed across multiple networks by accounting for an overlap between these models in a manner that reduces cross-network redundancy and increases the specificity of the network models for the network in which they are found. According to another aspect, a meta-network model is created for networks by accounting for the occurrence of network models that are found in multiple networks in a manner that reduces the redundancy across networks and that increases the specificity of the network model score. Preferably, this process provides additional weighting factors for each node in the network model.

1 . A method to determine a score for a degree of activation of an original network model, the original network model composed of causal connections among a set of nodes that represent biological entities, wherein nodes used to compute the score for the network have associated weights, wherein the score for the network is computed from scores for nodes in the network, and wherein the score for each node is computed from a set of measurable nodes, comprising:
 defining, and holding in a computer memory, one or more additional network models that are each comprised of a source node from the original network model connected to a set of measurable downstream nodes via causal edges; and   scoring the original network model using one of: modified network node weights, and modified additional networks linking network nodes to downstream measurables, to increase a specificity of the score for the original network model;   wherein the scoring is carried out in software executing in a hardware element.   
     
     
         2 . The method as described in  claim 1  wherein only a subset of the measurable downstream nodes is used for scoring, and wherein the subset of measurable downstream nodes is identified by evaluating an experimental context in which a relationship between the source node and at least one downstream measurable node is discovered. 
     
     
         3 . The method as described in  claim 1  wherein only a subset of the measurable downstream nodes is used for scoring, and wherein the subset of measurable downstream nodes is identified by determining which of one or more of the measurable downstream nodes are known to be associated with a biological process described by the original network and using at least one of those measurable downstream nodes in the subset. 
     
     
         4 . The method as described in  claim 1  wherein only a subset of the measurable downstream nodes is used for scoring, and wherein the subset of measurable downstream nodes is determined by removing from the full set one or more downstream measurable nodes that are found in less than a minimum percentage of the networks linking nodes in the original network to downstream measurables. 
     
     
         5 . The method as described in  claim 4  wherein only a subset of the measurable downstream nodes is used for scoring, and wherein the subset is computed by determining a fraction of node scores in the original network in which each measurable downstream node also contributes, setting to zero the weights for measurable downstream nodes whose fractional support is less than some configurable minimum, and further normalizing the remaining weights for that measurable downstream node to sum to a given value. 
     
     
         6 . The method as described in  claim 1  wherein only a subset of the measurable downstream nodes is used for scoring, and wherein the subset of measurable downstream nodes is determined by evaluating a relative context of measurable downstream nodes among the network models. 
     
     
         7 . The method as described in  claim 6  wherein the subset is computed by:
 determining a fraction of nodes in a network in which that measurable downstream node also contributes; 
 computing a fractional support for that measurable downstream node in each other network; 
 if the fractional support for that measurable in that network is less than the fractional support for that measurable in any other network, setting the measurable weight for that node to zero in that network; and 
 normalizing remaining measurable weights for each node in that network to sum to a given value. 
 
     
     
         8 . The method as described in  claim 1  wherein the network score is determined by weighting a contribution of each measurable downstream node to the node scores in each network based on the number of node scores in that network and in other networks to which each measurable contributes. 
     
     
         9 . The method as described in  claim 8  wherein the subset is determined by:
 for each measurable used to compute a node score, computing a promiscuity weight as the number of nodes in that network in which that downstream measurable node also contributes, divided by the number of nodes in which that downstream measurable node contributes across all networks including that network; 
 computing a new weight for that measurable for that node by multiplying an original measurable weight by the promiscuity weight, and re-normalizing all the measurable weights for each node in that network to sum to a given value.