Designing Efficient Multi-Objective Optimization Algorithms for Personalized Cancer Therapy

In this article, we present a new multi-objective optimization algorithm called DosaMO, designed to solve personalized regression problems. DosaMO takes into account the complexity of real-world decision-making processes by incorporating multiple objectives and considering adjusted performance. The proposed algorithm is based on the NSGA-II framework but with added capabilities for personalizing the tournament strategy, sorting algorithm, and objective functions.
To tackle the problem of adjusting regression models, we propose a two-stage approach. In the first stage, a second MO optimizer is deployed to generate a solution set by adjusting the performance of each objective using pre-trained regression models. Each original objective function is replaced by a pipeline that computes the function’s result, standard deviation, and feature count of the solution. The data is then fed to the corresponding adjuster regressor to produce the results for the user.
In the second stage, we train an adjuster learner using the original fitnesses, standard deviations, and features of each solution. We use a trainAdjuster function to create the adjuster learner, which is then used to adjust the objectives by training a new regression model for each objective. The resulting adjusted objectives are then fed to the main optimizer, which is also an MO optimizer that uses these adjusted objectives to produce the final results.
We evaluate DosaMO through experimental validation on a real-world dataset and compare it with other state-of-the-art algorithms. Our results show that DosaMO outperforms other algorithms in terms of computational efficiency, solution quality, and adaptability to different problem settings.
In summary, DosaMO is a powerful tool for solving personalized regression problems. By incorporating multiple objectives and adjusting the performance of each objective using pre-trained regression models, DosaMO can produce high-quality solutions that take into account the complexity of real-world decision-making processes. With its ability to adapt to different problem settings and efficient computational time, DosaMO is an excellent choice for practitioners and researchers alike.