Closed:RNA-Seq Best way to handle multimapping

Question

[firestar]

Hi, I am using STAR to align my illumina reads to a reference. For a multimapped read that maps equally well to multiple locations, should I

1. keep all alignments for multiple locations
2. keep one alignment (one location, primary alignment) and discard all other alignments (secondary alignments)
3. remove all alignments (primary and secondary alignments) for all locations

Which is better (1,2,3) for differential expression analysis [FEATURECOUNTS+DESEQ2] and differential isoform analysis [RSEM]? Should I filter first by MAPQ and then by primary alignment? Also how does one filter a BAM file for primary alignments?

Is it better to change some setting with the aligner during mapping or with the DGE tool?

Thanks.