Predicting adverse events accurately represents a significant challenge to both the pharmaceutical industry and academia, the reason being that our existing knowledge of biology, disease mechanisms (i.e. how a disease affects the healthy state of a human) and drug design is incomplete and sometimes incorrect.
In silico models are usually developed by extracting interactions and behaviors of biological systems either from the literature or from experimental data on a specific disease or biological system and integrating this information in some kind of a mathematical model that can be used to understand and predict the behavior of a drug in an organism.
Another relatively recent method is based on mining the scientific literature and correlating evidence from seemingly unrelated drugs or medical conditions.
While in silico methods aim to capture in depth the current knowledge of a biological system or a disease mechanism, they are still subject to the accuracy of that knowledge and may miss information that while seemingly unrelated, could in a multiply interconnected complex biological system prove highly relevant.
This gap is addressed by the literature-based discovery approach which does not capture details to the same extent but compensates by offering complete coverage of the available knowledge from all potentially related fields.