Does the weights file stay the same for a different model organism

Question

[Sharma]

I am making a new ML model for genomics based off an existing model. Key difference: model organism. The old model uses bigger, human fastq files, whereas mine will be made for yeast data. Should I make a new weights file for training the new model? or use the old one

PS I am new to computational biology and ML

[dthorbur]

You'll have to provide more information with this question. What does your model predict? If it's a process that is similar between human and yeast (e.g., protein structure predicted from peptide sequence is generally considered to be consistent between evolutionarily distinct groups), you can probably keep the old weights. If it's influenced by less conserved aspects (e.g., domain frequency, type and sequence of transcriptional elements, etc... ), you will probably need to train a new model.

With all of that said, you can always test performance of your model on data from a different organism. That seems like a reasonable starting point.