To Be or Not To be: My Take on Mechanistic vs Data-Driven Modeling

 My entire PhD (that's about 4 years of my life) was dedicated towards understanding the human innate immune system using computational modeling. I worked closely with an immunologist and a computational biologist to build this model. Having a Physics background, it was natural for me to delve into the mechanisms of each of the biological entities in the system I was working with, which I fondly called HIIS (human innate immune system), and tackled modeling using the mechanistic approach. That is, I described each and every biological mechanism as an equation coupled with the right parameters that go with it, and thus the entire system as a bunch of coupled ordinary differential equations. Hence, the birth of the HIIS model.

I mean, just look at it. 

   Presbitero, A., Mancini, E., Brands, R., Krzhizhanovskaya, V. V., & Sloot, P. M. A. (2018). SupplementedAlkaline Phosphatase Supports the Immune Response in Patients Undergoing Cardiac   Surgery: Clinical and Computational Evidence. Frontiers in Immunology, 9, 2342. https://doi.org/10.3389/fimmu.2018.02342

More importantly, as I was dealing with immunologists and biologists who demand for interpretability of the model itself, and not just the results, it calls for a mechanistic approach 

The data-driven approach on the other hand, much like what its name suggests, requires... you guessed it, DATA. In contrast with the mechanical approach, which establishes a bunch of equations based on known mechanisms (usually found in literature) and therefore does not really require data, in the data-driven approach, however, a model is used to describe the data at hand. A simple example would be describing the trend of a data using regression. 

Working as a data scientist allowed me to delve deeper (and swim about) into the data-driven approach. I use machine learning to be able to make sense of data, and predict the outcomes of certain scenarios that are relevant to the problem at hand based on the data that I have. 

Both have their pros and cons. 

The bottom line is, it all boils down to what you really want, and how you really want to make sense of your results. More importantly, what the stakeholders demand from you. I'm having fun with both!