Comparative study of Federated Learning and Machine Learning for Drug Discovery

It is widely recognized that machine learning has proven to be one of the most effective technologies in the field of bioinformatics, particularly in drug discovery. However, this technology is associated with certain challenges, including data security concerns, the risk of exposing patients’ personal medical information to malicious actors, and the costs involved in transferring data to a central repository. These challenges can be addressed using a novel technology called Federated Learning, which trains models on local devices or locations by sharing only the learned parameters with a global model, rather than the actual data. In this study, we apply Federated Learning in drug discovery and conduct a comparative analysis between Machine Learning and Federated Learning to determine if the results achieved by Federated Learning are comparable to those of traditional Machine Learning. While numerous studies have been conducted to compare machine learning algorithms, there has been limited comparison with federated learning. Through this analysis, we aim to identify the differences between the two technologies and determine whether federated learning achieves the same, better, or worse results. In our analysis, we obtained slightly better results, thereby achieving our objective of securing data, protecting patient privacy, and eliminating data transfer costs using Federated Learning for drug discovery. In this study, we implemented six algorithms—Artificial Neural Networks, Logistic Regression, Perceptron, Ridge Classifier, Ridge Classifier with Cross Validation, and SGD Classifier—for comparative analysis in both Federated Learning and Machine Learning. Consequently, we have been accomplished a comparison of both technologies.