Somatic mutation calling on tumor biopsies coming from cancer patients finds an increased foot hold in clinics all over Switzerland. The results of these tests aiming at rationalizing the use of certain cancer drugs, e.g., using Vemurafenib in case of a V600E mutation in melanoma, or in some cases indicating a lack of benefit for a drug, e.g., an NRAS mutation is predictive for the resistance of colon cancer against Cetuximab. While these examples are well studied and thus the causal link between the mutations and the effect of their respective drug is identified, the process of inferring the use of a certain drug with respect to a given mutation inside the genomic landscape of a tumor remains very complicated.
Identifying the presence of a mutation in the genomic landscape of a tumor is also a complicated task, but in contrast to finding the right drug for a given genomic variant, it can be done using algorithms enabling involved technical personnel to focus their time on quality control. When a list of mutations has been put together, an interdiscilinary team of bioinformaticians, pharmacogeneticists, molecular biologists and oncologists has all their work still ahead of them. They will spend many hours on finding useful drug-gene interactions, assessing the scientific literature for their level of evidence, classifying drugs in potentially beneficial or ineffective, and finally also cross-linking study populations with the clinical data presented by the individual patient. Only then it is possible to make an informed suggestion based on molecular evidence.
Or isn't it...
A lot of the time of this team will be spent on manually searching specific mutations in databases for drug-gene interactions, typing names of mutated genes and drugs in PubMed search boxes, looking up clinical trial opportunities and reading irrelevant scientific publications from a huge pool of cancer literature.
However, the information (not data) required to work this process is all out there. Neatly packed in databases and XML dumps, accessible using command line queries in an automated fashion. For instance the Drug Gene Interaction database combines information from 15 other databases on which drugs might target a certain gene and the functional requirement a mutation has to carry, for the drug to work. Repositories of clinical trials can be searched in the same way, e.g. NIH's clinicaltrials.gov. Abstracts and ultimately whole papers can be indexed using text miners quantifying their relevance according mutational and pharmacological information - like Google orders your websites according to relevance.
Do not get me wrong! This will not replace the expertise of the aforementioned team of experts, making a final decision on therapy suggestions. But it will make their work much more efficient; replacing sequential tasks like querying a database with an algorithm, selecting a set of relevant papers using text mining instead of clicking and skimming a text for some buzz words, and so on.
This will ultimately lead to better informed and faster decisions taking into account all available sources.
