A Personalized Cancer Drug Recommendation System with Machine Learning
Jianyu Liu, Yu He, Junfeng Li · Research Square · 2024
Abstract The patient’s gene sequence provides valuable information for clinical medication. In practice, there are many different types of data on genes but they are not fully utilized. There are different mutation genes in different cases of cancer, in which there are correlations between them. The gene-based drug recommendation system can be used as an auxiliary tool for clinicians. However, the performance of existing models is not up to the standard of clinical application. In this context, we intend to propose a better-performance drug recommendation system.We propose a multi-kernel ranking learning drug recommendation model, which uses multi-kernel learning methods to extract features from multimodal data and to rank drugs by learning the correlation between different data. Our model achieves performance improvements, especially in the case of simulated clinical sparsity; also has the flexibility to incorporate more multimodal data, which can further optimize the model for potential translation to clinical applications. Despite the limited data used to train and test regression models, model performance was found to be improved. MKRL model can use multimodal data to continuously add to the model for training, and our drug recommendation effect is better than other recommendation models.