There are many choices and assumptions to make when designing a machine learning (ML) based system. Taking the common choice is appealing but can undermine your system performance. Having recently designed an ML based system for prediction of gene expression (GE) , we made various uncommon but sensible choices and assumptions given the particular problem we solved. I’d like to highlight some of those choices here and elaborate why they are sensible. Figure 1. briefly introduces our GE prediction problem where we want our model to use expression of master regulator (MR) gene and knockout vector to predict GE values for other genes.
One of the concepts that can improve effectiveness of a machine learning (ML) method, is the consideration of sparsity in its design. Here I give a short summary on benefits of sparsity considerations in ML.