Inconsistency in reproducing tensorQTL output using R lm() function

[james-li-projects]

Hello, I had utilized tensorQTL to successfully map cell type interaction eQTLs. However, when I tried to manually run linear regressions for these interaction QTLs using R code using this setup: GeneExpresssion ~ SNP + CellType + SNP*CellType + covariates, I unfortunately obtained different effect size estimates and p-values for the interaction term and main effects. Just for context, I double-checked that the joined table (with gene expression, variant dosages, estimated/normalized cell type scores, and covariates) was properly joined. These are some differing results I obtained:

From tensorQTL:
pval_gi=9.26011e-07
b_gi=-0.195591

From R's lm() function:
pval_gi=6.072221e-06
b_gi=-0.1740634

I was a bit surprised to see this difference since I had originally assumed tensorQTL may be mapping interaction QTLs based on the same linear regression setup. Do you by any chance know if tensorQTL may be adding or removing additional terms from this regression, performing additional filtering of samples, or if you may have any suggestions on how I could potentially modify my regression setup in R to match tensorQTL’s setup? Any help would be immensely appreciated! Thank you.