Predicting the Adverse Reaction of Drugs by analyzing their molecular interactions within the body
Drugs are designed to cure diseases but unfortunately they also tend to cause a multitude of undesired effects in the human body called adverse drug reactions (ADRs). Formally defined as a "response to a drug which is noxious and unintended and which occurs at doses normally used in man for prophylaxis, diagnosis, or therapy of disease or for the modification of physiologic function", ADRs have recently given rise to widespread public concern. Every year approximately 2 million patients in the United States are estimated to be affected by a severe ADR resulting in roughly 100,000 fatalities. This makes ADR the fourth leading cause of death in the U.S, right after cancer and heart diseases. ADRs are estimated to account for about 2.5% of all hospitalized cases and this consequently places a huge burden on the national economy. It is estimated that $136 billion is spent on treating ADRs in the U.S. every year and other nations worldwide are also facing similar difficulties. The incidence of several serious and unexpected ADRs led to the withdrawal of 19 widely used marketed drugs over the last decade making ADR a major commercial concern for pharmaceutical companies. Moreover, the increase in safety requirements due to the prevalence of ADRs have raised the costs and slowed down the speed of the drug development process . Thus, developing effective mechanisms for detecting potential ADRs before marketing drugs is now critically important. The tools for doing this are currently very limited and extensive research needs to be done to make this task more efficient, inexpensive and reliable.
One promising means for predicting the adverse effects of a drug involves the study of its molecular interactions within the human body. Drugs exert their effects on the body through these interactions and analyzing these interactions should provide useful insight into the causes of the known adverse reactions of these drugs, thereby facilitating the detection of unknown ADRs. By using advanced text mining techniques for extraction of all relevant data from biomedical literature, this process can prove to be cheaper and more informative than in vitro Safety Pharmacology Profiling. Researchers have already started looking into this new direction, but it still remains a largely unexplored region.
Tools and Methods:
GeneRanker is a tool for ranking genes associated with a given disease by mining the huge volume of biomedical data available in the CBioC database. By extending GeneRanker to mine ADR-gene relationships and drug-gene relationships with the help of PTQL, it can be used to study links between drugs and their known ADRs. First, the GeneRanker can be used to test the hypothesis that there is a significant similarity between the network of proteins affected by a given drug and the protein network involved in its known ADRs. To do this, the top ranked genes for a given drug can be compared with those for its known ADRs and for random ADRs. If the hypothesis is true then the degree of overlap between the target genes of the drug and the genes associated with the known ADRs should be notably greater than that between the former and the genes associated with random ADRs. Some genes that tend to be common in most diseases may be filtered out during this step to make the difference between the two results stronger.
Once the given hypothesis has been established, more sophisticated techniques for comparing the protein network affected by a drug with that associated with its ADRs may be employed to gain further insight. Network comparison is a computationally intensive process and tools for doing so are still not adequately advanced. There are mainly three modes of comparative methods: network alignment, network integration and network query. Network alignment entails global comparison of two different networks by identifying areas of similarity and dissimilarity. Network integration is the process of merging different networks together and studying how they are interconnected. Network query involves finding subnetworks within a network that are similar to a subnetwork of interest. A complete comparison of two different networks is infeasible since this will require solving the NP hard problem of subgraph polymorphism. Thus special heuristics are used to allow relevant comparison of large biological networks [10, 11, 12]. There are a few tools available online for this purpose. These include Cytoscape, NeAT, MetNetAligner, Graemlin etc.
For this project, we may start by using some of the freely available tools for network analysis to compare the drug-protein network and ADR-protein network and then attempt to build our own system for network comparison. Such comparison may prove to be important in several ways. For instance, Jian Yang is building a web crawler to extract all mention of ADRs associated with a given drug in social networks. Through our analysis we may decide on a threshold level of similarity between the ADR-protein network and drug-protein network for filtering out the false positives. More detailed analysis can provide further information. If we can compare the regions of similarity between a given ADR-protein network and the protein network of different drugs known to cause the ADR, we may be able to infer which proteins in the network are more likely to be the cause of this ADR. Other ADRs that are known to interact with these proteins are also likely to be the ADRs of these drugs. We may also compare the protein networks involved in different drugs to determine which drugs are likely to interact and produce an adverse reaction.
1. Use gene ranker to mine ADR-gene and drug-gene relationships
2. Determine the degree of similarity between ADR-gene and drug-gene networks by using available online software
3. Choose a threshold level of similarity to filter out false positives from the list of ADRs gathered by Jian Yang
4. Perform more complex analysis of the networks to determine the causes of known ADRs of drugs and evaluate the methods used
5. Submit paper discussing the results