Dear @ryzhu,
Check with FASTQC if there is any adapter contamination still in your data. Furthermore, it might be worth to extract your multi-mapped reads and use featurecounts or blast to see if these come from rRNA (using a gff/gtf for featurecounts or a rRNA sequence file for blast). High ratio of muli-mapped reads in RNA-seq is often a sign of an incomplete depletion of rRNA.
Kind regards,
Florian