During 1-1 meeting today, the idea of LDFI 4 life was discussed a bit more in the context of potential upcoming projects. In order to articulate the idea fairly well, the conversation went a little something like this.
the “LDFI approach” is composed of roughly 3 components, each of which should have some mapping to the domain to be studied:
observability/tracing - on some level this could be using fluorescence to track where a protein may be present/visible, or it could be measuring expression in certain cells, or it could be quantity of cell-free molecules, or it could even be results of a simulation
fault-injection or what to perturb in order to get a different result - this could be a suggestion to knockout a different gene, this could be to make some modification of a protein or gene, or this could be a different string representation to use for dna, or different transcript to use, all in order to prevent a signal from one subsystem/pathway to another (where proteins are cellular signals).
there are multiple “loops” in biological science:
wetlab/experimental - scientists conduct experiments to study things such as gene knockouts, then based on results they alter the type of experiment they are interested in, or what they are testing in the experiment. For genomic experiments, this may be to research new protocols, or to find what differentiates response between multiple patients
bioinformatics/genomics - once an experiment is done, and the data is assembled or propagated up to higher level analysis, higher dimensional analysis or comparison to other experiments in a more computational matter is done, and then the experiment is altered in similar ways as mentioned above
there are many other loops, each dealing with different levels of abstraction or different levels of analysis in which some information is available which is not available at the abstraction above or below.
in each “loop,” the LDFI approach can make different choices as to what interesting experiments can or should be done
Lineage driven fault-injection was used to make testing of systems more efficient by avoiding faults that were known to be problematic and by avoiding faults that the system was known to tolerate.