Merging peaks for sample comparisons

[samfenske]

Hello,

Thank you for the excellent resource and documentation! I have a question regarding experimental design. I am interested in using ChromBPNet to infer differential TF binding motifs between cases and controls. I had previously trained individual models on each sample, but realize that this leads to inconsistencies in comparing feature importance scores across different trained models. For a more robust comparison, would you recommend merging peaks across all samples and training a single model and then evaluating on individual samples to get feature importance scores and seqlets/motifs?

Thanks!