Stringtie inconsistency

[kuristy]

I am doing a large RNAseq experiment, and I am doing hisat2-stringtie-ballgown as we get new samples. After alignment and assembly, I do stringtie --merge with new samples and older samples together. I then run stringtie -eB to generate the ballgown tables for all of the samples old and new. We noticed that **only for some genes**, the fpkm values will change after more samples are added. And, as more samples are added, the number of genes that show varying fpkms increases. For example, after adding 10 new samples I have ~50 genes that changed by at least 15 fpkm for **the same sample** that was run previously. After adding 20 samples I have almost 200 genes that changed by at least 15 fpkm for the same sample. I am using the exact same bam files from run to run. The only thing that changes is the merge file after adding new samples. I realize stringtie is not 100% error free, but it seems strange that adding more samples would cause a change in fpkm for only some of the genes, and that adding more samples causes even more genes to change from the same sample.

Here are my commands:
stringtie -M 0.75 -f 0.3 -c 3.5 -j 10 -m 400 --rf "${base}"_sorted.bam -p 4 -G /media/manager/Larkins_1TB/indexes/grch38_snp_tran_ht2/gencode.v27.chr_patch_hapl_scaff.annotation.gtf -o /media/manager/Larkins_1TB/stringtie/"${base}".gtf

stringtie --merge -p 8 -f 0.3 -c 3.5 -m 400 --rf -G /media/manager/Larkins_1TB/indexes/grch38_snp_tran_ht2/gencode.v27.chr_patch_hapl_scaff.annotation.gtf -o stringtie_merged.gtf mergelist.txt

stringtie /media/manager/Larkins_1TB/"${base}"_sorted.bam -M 0.75 -f 0.3 -c 3.5 -j 10 -m 400 --rf -eB -p 8 -G stringtie_merged.gtf -o "${base}"/"${base}".gtf

Let me know if there is anything I can change to get consistency from run to run after new samples are added to experiment.


Thanks!