Limits...
Enhanced stochastic optimization algorithm for finding effective multi-target therapeutics.

Yoon BJ - BMC Bioinformatics (2011)

Bottom Line: However, biological networks are typically robust to external perturbations, making it difficult to beneficially alter the network dynamics by controlling a single target.In fact, multi-target therapeutics is often more effective compared to monotherapies, and combinatory drugs are commonly used these days for treating various diseases.A practical challenge in combination therapy is that the number of possible drug combinations increases exponentially, which makes the prediction of the optimal drug combination a difficult combinatorial optimization problem.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX 77843-3128, USA. bjyoon@ece.tamu.edu

ABSTRACT

Background: For treating a complex disease such as cancer, we need effective means to control the biological network that underlies the disease. However, biological networks are typically robust to external perturbations, making it difficult to beneficially alter the network dynamics by controlling a single target. In fact, multi-target therapeutics is often more effective compared to monotherapies, and combinatory drugs are commonly used these days for treating various diseases. A practical challenge in combination therapy is that the number of possible drug combinations increases exponentially, which makes the prediction of the optimal drug combination a difficult combinatorial optimization problem. Recently, a stochastic optimization algorithm called the Gur Game algorithm was proposed for drug optimization, which was shown to be very efficient in finding potent drug combinations.

Results: In this paper, we propose a novel stochastic optimization algorithm that can be used for effective optimization of combinatory drugs. The proposed algorithm analyzes how the concentration change of a specific drug affects the overall drug response, thereby making an informed guess on how the concentration should be updated to improve the drug response. We evaluated the performance of the proposed algorithm based on various drug response functions, and compared it with the Gur Game algorithm.

Conclusions: Numerical experiments clearly show that the proposed algorithm significantly outperforms the original Gur Game algorithm, in terms of reliability and efficiency. This enhanced optimization algorithm can provide an effective framework for identifying potent drug combinations that lead to optimal drug response.

Show MeSH

Related in: MedlinePlus

Updating the drug concentrations. (A) Finite state automaton for sequentially updating the two drugs. Each node corresponds to a specific drug combination. States that have the same concentration for drug 1 are aligned in the same column. Similarly, states with the same concentration for drug 2 are aligned in the same row. The arrows show the allowed transitions between sates. (B) Finite state automaton for simultaneously updating the two drugs.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3044272&req=5

Figure 5: Updating the drug concentrations. (A) Finite state automaton for sequentially updating the two drugs. Each node corresponds to a specific drug combination. States that have the same concentration for drug 1 are aligned in the same column. Similarly, states with the same concentration for drug 2 are aligned in the same row. The arrows show the allowed transitions between sates. (B) Finite state automaton for simultaneously updating the two drugs.

Mentions: For each of the four response functions shown in Figure 4, we tested the performance of the proposed algorithm as follows. First, we randomly selected the initial values of x and y (i.e., initial drug concentrations). Next, starting from the selected initial values, we used the proposed algorithm to search for the optimal drug combination (x, y) that maximizes the drug response. The parameter α, which is used for controlling the randomness of the search, was set to α = 1. In every experiment, we continued the search for 4NxNy iterations, where Nx is the number of distinct concentrations for x and Ny is the number of distinct concentrations for y. To obtain a reliable performance estimate, this experiment was repeated 10,000 times. Based on the 10,000 independent experiments, we estimated the success rate S, which is defined as the relative number of experiments, in which the algorithm was able to find an effective optimal drug combination (x, y) within NxNy (i.e., total number of distinct drug combinations) iterations. We consider a combination (x, y) to be effective if f(x, y) ≥ λ for a given λ ∈ [0,1], or if the combination (x, y) is among the top P% combinations that result in the highest drug response. In addition to the success rate, we also estimated the average number iterations that were needed to find an effective drug combination, in case the experiment was successful. We also performed similar experiments using the Gur Game algorithm, to compare the performance of the two algorithms. Since the Gur Game algorithm does not make use of the drug response change that results from the concentration change of a specific drug, the two drug concentrations x and y can be either updated simultaneously or sequentially (one after the other). Sequentially updating the two drugs corresponds to using the FSA shown in Figure 5A, while updating them simultaneously corresponds to using the FSA illustrated in Figure 5B. As before, the Gur Game algorithm was designed such that it determines the direction of reward by comparing the current drug concentration to the central concentration. We evaluated the performance of the Gur Game algorithm based on the simultaneous update approach as well as the sequential update approach.


Enhanced stochastic optimization algorithm for finding effective multi-target therapeutics.

Yoon BJ - BMC Bioinformatics (2011)

Updating the drug concentrations. (A) Finite state automaton for sequentially updating the two drugs. Each node corresponds to a specific drug combination. States that have the same concentration for drug 1 are aligned in the same column. Similarly, states with the same concentration for drug 2 are aligned in the same row. The arrows show the allowed transitions between sates. (B) Finite state automaton for simultaneously updating the two drugs.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3044272&req=5

Figure 5: Updating the drug concentrations. (A) Finite state automaton for sequentially updating the two drugs. Each node corresponds to a specific drug combination. States that have the same concentration for drug 1 are aligned in the same column. Similarly, states with the same concentration for drug 2 are aligned in the same row. The arrows show the allowed transitions between sates. (B) Finite state automaton for simultaneously updating the two drugs.
Mentions: For each of the four response functions shown in Figure 4, we tested the performance of the proposed algorithm as follows. First, we randomly selected the initial values of x and y (i.e., initial drug concentrations). Next, starting from the selected initial values, we used the proposed algorithm to search for the optimal drug combination (x, y) that maximizes the drug response. The parameter α, which is used for controlling the randomness of the search, was set to α = 1. In every experiment, we continued the search for 4NxNy iterations, where Nx is the number of distinct concentrations for x and Ny is the number of distinct concentrations for y. To obtain a reliable performance estimate, this experiment was repeated 10,000 times. Based on the 10,000 independent experiments, we estimated the success rate S, which is defined as the relative number of experiments, in which the algorithm was able to find an effective optimal drug combination (x, y) within NxNy (i.e., total number of distinct drug combinations) iterations. We consider a combination (x, y) to be effective if f(x, y) ≥ λ for a given λ ∈ [0,1], or if the combination (x, y) is among the top P% combinations that result in the highest drug response. In addition to the success rate, we also estimated the average number iterations that were needed to find an effective drug combination, in case the experiment was successful. We also performed similar experiments using the Gur Game algorithm, to compare the performance of the two algorithms. Since the Gur Game algorithm does not make use of the drug response change that results from the concentration change of a specific drug, the two drug concentrations x and y can be either updated simultaneously or sequentially (one after the other). Sequentially updating the two drugs corresponds to using the FSA shown in Figure 5A, while updating them simultaneously corresponds to using the FSA illustrated in Figure 5B. As before, the Gur Game algorithm was designed such that it determines the direction of reward by comparing the current drug concentration to the central concentration. We evaluated the performance of the Gur Game algorithm based on the simultaneous update approach as well as the sequential update approach.

Bottom Line: However, biological networks are typically robust to external perturbations, making it difficult to beneficially alter the network dynamics by controlling a single target.In fact, multi-target therapeutics is often more effective compared to monotherapies, and combinatory drugs are commonly used these days for treating various diseases.A practical challenge in combination therapy is that the number of possible drug combinations increases exponentially, which makes the prediction of the optimal drug combination a difficult combinatorial optimization problem.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX 77843-3128, USA. bjyoon@ece.tamu.edu

ABSTRACT

Background: For treating a complex disease such as cancer, we need effective means to control the biological network that underlies the disease. However, biological networks are typically robust to external perturbations, making it difficult to beneficially alter the network dynamics by controlling a single target. In fact, multi-target therapeutics is often more effective compared to monotherapies, and combinatory drugs are commonly used these days for treating various diseases. A practical challenge in combination therapy is that the number of possible drug combinations increases exponentially, which makes the prediction of the optimal drug combination a difficult combinatorial optimization problem. Recently, a stochastic optimization algorithm called the Gur Game algorithm was proposed for drug optimization, which was shown to be very efficient in finding potent drug combinations.

Results: In this paper, we propose a novel stochastic optimization algorithm that can be used for effective optimization of combinatory drugs. The proposed algorithm analyzes how the concentration change of a specific drug affects the overall drug response, thereby making an informed guess on how the concentration should be updated to improve the drug response. We evaluated the performance of the proposed algorithm based on various drug response functions, and compared it with the Gur Game algorithm.

Conclusions: Numerical experiments clearly show that the proposed algorithm significantly outperforms the original Gur Game algorithm, in terms of reliability and efficiency. This enhanced optimization algorithm can provide an effective framework for identifying potent drug combinations that lead to optimal drug response.

Show MeSH
Related in: MedlinePlus