Table 1. Examples of misconceptions versus the intended use accepted by experts in the field.

Misconceptions or misleading statements	Correct meaning
Computational methods are fast and cheap, and in particular in the pandemic situation we can, and should, put experimental research on hold and turn to in silico methods.	In silico studies may be conducted independently of experiments. In fact, theoretical approaches may address research questions that are beyond experimental accessibility. Ideally, computation and measurement are integrated, e.g. for the purpose of model validation and in applied research.
Computational studies are fast and easy to conduct, and they always produce results.	Depending on the research question, computational projects can in fact be resource-demanding and time-consuming. The fact that they “always” produce results does not mean that these results are “always” valid (this is certainly not the case). Use of any results without careful vetting is related to a significant risk of predictions being false or inaccurate.
Purely computational studies have limited value and do not represent standalone projects.	A purely computational study can be self-contained and comprehensive and may address research questions going beyond experimental accessibility.
In multidisciplinary projects, experimental testing is difficult but computational studies are easy.	Both computational and experimental work might be routine or challenging. The development and validation of new algorithms may well exceed the magnitude of experimental work.
Computational analysis mostly contributes catchy pictures to publications and grant applications.	If properly conducted, computational analysis can rationalize experimental observations and yield experimentally testable hypotheses.
Machine learning and AI are the new standard for CADD.	Machine learning is a part of AI and already has a long record of use in CADD. While being important to many types of predictions and enabling new applications, machine learning methods on their own have not revolutionized the field (yet).
Molecular modeling and chemoinformatics are other terms for CADD.	CADD covers various theoretical disciplines including molecular modeling, chemoinformatics, bioinformatics, theoretical chemistry, and machine learning.
Molecular docking can be used to demonstrate ligand binding.	Molecular docking approximates protein-ligand interactions and binding modes in a computational complex manner that only partly resembles physical binding events. Entropic effects in particular are only poorly considered by docking approaches.
Rational drug design must incorporate a computational analysis.	A drug can be rationally designed based on prior knowledge, experience and even causal intuition, without the need to employ a computer.
Computational results are unbiased and thus most likely correct.	Any computational analysis is affected by methodological limitations in accounting for the physical reality. Hence, results must be interpreted with caution and awareness of such limitations. It is ultimately the responsibility of the researcher to arrive at a scientifically sound and trustworthy interpretation of the results.
Most computational techniques can be learned in a few hours of hands-on workshops.	How to execute a software might be learned rapidly, but understanding the theory behind a computational approach and gaining the experience essential to the correct interpretation of the relevance, meaning and reliability of predictions can be demanding and take a significant amount of time. Without a firm grasp of the underlying theoretical foundations, computational exercises may only lead to appealing but meaningless illustrations.
In contrast to reagents for organic synthesis, no “purification” is needed for the input of computational approaches.	Data curation is of fundamental importance to CADD. Without proper curation of the input data, no meaningful predictions will be possible. The role of data curation in CADD equals that of experimental preparation and reagent purification in chemical synthesis.
In contrast to reaction mechanisms in organic synthesis, understanding of how an algorithm works is not required in order to produce predictions.	Understanding algorithms is as important as understanding experimental approaches.
In contrast to a biochemical assay, controls are not required in a computational analysis.	Including controls into the calculations is an indispensable part of any scientifically sound computational study.
There are no computational “experiments”.	An inferior computational analysis will produce results meeting our pre-formed expectations. A properly designed computational study will unambiguously address a question or hypothesis for which we have no answer to yet. This represents, in its best sense, an “in silico experiment”.