| Summary: | The current hype associated with machine learning and artificial intelligence often confuses scientists and students and may lead to uncritical or inappropriate applications of computational approaches. Even the field of computer-aided drug design (CADD) is not an exception. The situation is ambivalent. On one hand, more scientists are becoming aware of the benefits of learning from available data and are beginning to derive predictive models before designing experiments. However, on the other hand, easy accessibility of in silico tools comes at the risk of using them as “black boxes” without sufficient expert knowledge, leading to widespread misconceptions and problems. For example, results of computations may be taken at face value as “nothing but the truth” and data visualization may be used only to generate “pretty and colorful pictures”. Computational experts might come to the rescue and help to re-direct such efforts, for example, by guiding interested novices to conduct meaningful data analysis, make scientifically sound predictions, and communicate the findings in a rigorous manner. However, this is not always ensured. This contribution aims to encourage investigators entering the CADD arena to obtain adequate computational training, communicate or collaborate with experts, and become aware of the fundamentals of computational methods and their given limitations, beyond the hype. By its very nature, this Opinion is partly subjective and we do not attempt to provide a comprehensive guide to the best practices of CADD; instead, we wish to stimulate an open discussion within the scientific community and advocate rational rather than fashion-driven use of computational methods. We take advantage of the open peer-review culture of F1000Research such that reviewers and interested readers may engage in this discussion and obtain credits for their candid personal views and comments. We hope that this open discussion forum will contribute to shaping the future practice of CADD.
|
|---|
