Table 1.

Examples of potential applications of transfer learning methods in drug design.

Transfer learning approach	Concept	Application in drug design
Instance-based	Uses the same ML technique for modeling but apply some changes to the parameters of the target model.	Source and target datasets have the same endpoint (e.g., same molecular target) but the training data can be colleted at different experimental conditions.
Feature representation	Based on some mathematical transformations of data.	Source and target datasets have different but related endpoints, e.g., same classes of molecular target (kinases, nuclear receptors, proteases, etc.).
Parameters	It is assumed that both datasets share some properties.	Source and target datasets have the same or related endpoints.
Relational knowledge transfer	Based on technique for mapping the data in the target domain.	The endpoints of the source and target datasets are different but the domains (the independent variable in QSAR models) are related; e.g., cellular permeability and log P.