So many people wonder if clinical trials could be replaced by computational trials. This happened at least in one case in USA. There are very sophisticated software in the area of PK/PD, but they are based on very simple assumptions such as storage and diffusion of drugs through the biological compartments.

A new article on Nature[0] tells that it is possible to predict side effects using a recommender system. Basically the scientists built a matrix between drugs and their side effects and inferred probabilities of side effects, including side effects that were still not observed. The researchers claim their tool makes it possible to correlate these probabilities with physiology. The code is available [1].

I guess it would be simple to extend this idea to predict how a drug would be successful for a disease. This is actually the goal of any clinical trial. This could save billions of dollars in research in biotechs and academia while saving time and life of patients. It would also render animal testing obsolete.

We have databases such as Drugbank [2] linking drugs to diseases. It is quite trivial to make a matrix linking diseases to drugs. The questions are "how much trustworthy it is?" and "how it could demonstrate trustworthiness to non-engineers or non-mathematicians, specially to medical specialists".

For having any success, the issue of trust certainly need a lot of scrutiny.

Any thought?

[0] https://pubmed.ncbi.nlm.nih.gov/32917868/

[1] https://github.com/paccanarolab/Side-effect-Frequencies

[2] https://en.wikipedia.org/wiki/DrugBank