Seeking help with deep learning model developement

I am a Ph.D. student and I work on the prediction of splice sites from genomic sequences using deep learning.

I find myself facing a particular challenge when it comes to initiating the development of a deep learning model. The question of where to begin often perplexes me. Is it a matter of trial and error, or should the choice of architectures be guided by relevant factors influencing the prediction task? Upon an extensive review of existing models for splice site prediction, it has come to my attention that many of them lack a specific rationale for the selection of their architectures. This has led me to believe that the process of model selection can be somewhat arbitrary and time-consuming.

With this in mind, I kindly seek your expert guidance and advice. I would greatly appreciate your insights on how to streamline the overwhelming array of architecture and hyperparameter possibilities, are there some general guidlines to follow ?

Hello there,

My suggestion is to try start up with architectures similar to your application and could possibly modify as you refine your model if needed. Its mostly trial and error process but they have done some trials with it from which you can benefit for your own application.