I am tying to do differential expression (DE) analysis between different conditions. Now I have same samples sequenced at multiple lanes to improve coverage. I am wondering if it would be better to combine those reads from different lanes for quantification or quantify the reads separately before DE analysis. Having more samples for DE analysis would improve the statistical power I suppose.
There is no gold standard for this. You can quantify and quality control all runs separately and then sum up the counts from the technical replicates prior to feeding them into the DE analysis to make sure to technical problems occurred during the sequencing run, which is very rare in my experience. Still, do not treat these lane replicates as independent replicates in the DE analysis, as they are simply sequencing and not biological replicates. Technically this is of course possible but it would underestimate the sample dispersion and create significance where is none.
In general, running the same library on different lanes of the same kind of instrument does add not technical variation. Unless there was a problem with the instrument, you should combine the data from the same sample together, no matter how many lanes it's spread out on.