Predictions are a part of life, everyone wants to know about their future. It has a different meaning for those working in the different fields like rain forecast, drug toxicity. Here, my question is related to scientific predictions. I wants to know how much experimentalist believe in theoretical predictions. I am also interested to know from users whether these prediction methods are really worthy?
I am specifically interested in the field of bioinformatics, chemoinformatics, pharmacoinformatics and clinical informatics. Whether in silico predictions are really useful in designing your experiments. In your view what is the limitation of existing prediction methods in your area of research and how one can improve these prediction techniques.
In my experience, there are fields where predictions can already be useful. But they are only part of the process to form a hypothesis which then needs to be tested. Usually in the wet lab.
The limitation of most predictions are not understood or known links or parts, which are then not present in the model.
Hi,
There are cases were prediction is pretty reliable and relatively accurate such as the association of an some inherited mutation to a disease or cancer development.
These rely on single on low number of parameters which in high probability (or definite) will lead to the final outcome.
Other medical predictions probably need a much more complex predicting model and for sure today there are many unknown factors and possible interactions which will reduce reliability.
An example comes to mind about smoking, how many parameters are still unknown and need to come together to determine whether a certain smoker will acquire lung cancer or other smoking related disease in the future.
In these cases we are not there yet, but I am sure (hope) as science progress and our understanding grows we will get there some time in the future.
Regards.
I ask myself every day how much can I trust the result of this analysis. I think that almost all biological data has some degree of uncertainty. In the wetlab I think people are probably not as aware of this fact as one is in Bioinformatics for instance.
The way I think about it is that a conclusion or prediction is only as good as the underlying model. The model requires understanding of the system, so the worse our understanding the worse our prediction and our conclusions.
High-throughput data is great for increasing understanding by exploratory analysis to improve experiments and models in statistical analyses. So I think the informatics and wetlab methods need to be mixed more in projects, wetlab people should trust results that rest statistics and informaticians should become better at understanding the specific biological or other context.
From my point of view as a wet biologists, predictions are "easy", testing them is a lot of work. Conceptually it is really difficult for experimentalists to deal with confidence intervals, likelihoods and multiple models. "What do you mean you are 90% sure that this mutation will affect this function?", "I wanted one model of the protein, not multiple models". I think if a prediction method is thoroughly validated, gives clearly understandable results and in the best case some of the results are either trivial (every wet biologist would predict them too) or already confirmed, then the method is worth to be further explored.
There is a huge gap between experimental and computational biology, experimental biologist want practical, testable predictions and computational biologist are often quite unaware of the methods and amount of work associated with testing predictions (conversation with a small molecule computational biologist "so why don't you test the predictions on all receptors?", because it's not a script which you run 32 times instead of one time but at least 32 times more wet work).
Prediction sometimes means garbage in and garbage out, that depends on how accurate the data you used for the model. Prediction without real experimental test is nothing in general.
Predictions without experimental input has no meaning. Experimental values in support of prediction makes the situation more real.
- Badges
- Science topic
- Similar topics
- Medicine
- Oncology
- Cancer Research
- Cancer Biology