This is a subset of the project Genetic Circuit Design for Boston University
Class EC/BE522 - Computational Synthetic Biology for Engineers.
This repository is for the technical interview portion for the position of Software Engineer at the DAMP Lab at Boston University. In this readme I will detail my technical choices and design decision to demonstrate my proficiency in software engineering.
Assuming familiarity with the conda ecosystem:
git clone https://github.com/0x174/pegbox.gitcd pegboxconda env create -f environment.ymlconda activate pegbox-devpytest
Due to some technical difficulties with interacting with the remote endpoint, I've encapsulated most of the functionality in terms of optimization into testing files that show the general pattern for generating circuits and optimizing them.
I have also written out a new python3.6+ API for interacting with the Cello service, but it is untested due to the above factors.
Scipy offers a variety of minimization algorithms, each of which has strengths and weaknesses. I hedged my bets by creating a uniform interface that allowed the user to select which algorithm would best be suited to their problem.
My choice of using Python was primarily motivated by the scientific library ecosystem and development speed.
If I were to have more time and this was deployed less as a script and more as a remote service as some form of optimization for Cello input in the wild, I would move to a more statically typed language for execution speed, either through some compiled cython bindings, subprocess calls to compiled C/C++, or something more exotic like Julia if I'm really trying to wring performance out and I can trust that the interface is going to be fairly stable.
My challenge related to this project were ultimately related to understanding the problem, which really took two major forms:
-
Understanding the semi-analog state of repressors and how they interact together.
-
Learning about the benefits and drawbacks of optimization algorithms and correct selection thereof.