Facing small and biased data dilemma in drug discovery with federated learning

Abstract

Artificial intelligence (AI) models usually require large amounts of high quality training data, which is in striking contrast to the situation of small and biased data faced by current drug discovery pipelines. The concept of federated learning has been proposed to utilize distributed data from different sources without leaking sensitive information of these data. This emerging decentralized machine learning paradigm is expected to dramatically improve the success of AI-powered drug discovery. We here simulate the federated learning process with 7 aqueous solubility datasets from different sources, among which there are overlapping molecules with high or low biases in the recorded values. Beyond the benefit of gaining more data, we also demonstrate federated training has a regularization effect making it superior than centralized training on the pooled datasets with high biases. Further, federated model customization for each client can effectively help us deal with the highly biased data in drug discovery and achieve better generalization performance. Our work demonstrates the application of federated learning in predicting drug related properties, but also highlights its promising role in addressing the small data and biased data dilemma in drug discovery.

Federated Learning: A Survey on Enabling Technologies, Protocols, and Applications

“…Surprisingly, FL has also been used for drug discovery. At least two works, one by [161] and another by [162]. Both works cover FL's applicability for drug discovery, however, these two works tackle specific problems related to drug discovery and apply FL to solve them.…”

“…Both works cover FL's applicability for drug discovery, however, these two works tackle specific problems related to drug discovery and apply FL to solve them. In [161], the focus on incorporating FL for drug discovery deals with datasets with high biases. Biased data can be a huge problem for model training because the data itself can be skewed due to these biases.…”

Federated learning of molecular properties with graph neural networks in a heterogeneous setting

“…There are several works focusing on federated graph neural networks 9 , 39 , 40 , 41 , 42 , 43 , 44 and federated molecular-property prediction. 45 , 46 GraphFL applies MAML to improve the robustness of training. 43 The method in Xie et al.…”

Federated Learning - Methods, Applications and beyond

“…Other applications of FL include ranking browser history suggestions based on user-interactions [43], visual object detection [44], patient clustering to predict hospital stay time as well as mortality [45], drug discovery [46,47] and brain tumor segmentation [48,49,50]. The application fields of FL are quickly increasing and are a promising research direction of ML.…”