Do I need to merge bam files or collapse replicates? I am working on differential expression analysis and have 3 technical replicates for each sample from different library not lane. Is it recommended to merge bam files? my question is that if the library sizes very different to be merged or collapse the read counts? will such estimate be true to the expression levels. How to move ahead any suggestions

I usually look at the samples separately by MultiQC (or whatever you prefer to QC of the fastq files) and then with plotPCA in DESeq2. If they appear as technical replicates I merged the files at the top of my pipeline. You can also sum the counts together in R using collapseReplicates in DESeq2.