GWAS Effect Sizes

First, yes if you swap the effect allele from ALT to REF, the effect sizes simply flip signs. If an ALT allele has a weight of +0.2, the REF allele for that same variant has a weight of -0.2.

PRS Computation

In a superficial numerical sense it is true that the number changes, but the two approaches still yield the exact same predictive information and mean the same thing, in particular after standardization (as below).

PRS tend to expect 2 alleles at any standard autosomal locus. Therefore, a person's REF dosage plus their ALT dosage always equals 2. Because of this fixed relationship, if you run the calculation using the REF weights and REF dosages, you obtain:

Ref-based score = ALT-based score - (2 * the sum of all the ALT weights across the scoring panel)

The key takeaway is that this constant shift is exactly the same for every individual in your dataset. Therefore, while distribution of scores shifts up or down the number line, it does so equivalently for everyone, and thus the relative distance between individuals never changes. If Patient A has a higher genetic risk than Patient B using the ALT score, they will still have a higher risk using the REF score.

Furthermore, if the polygenic scores is mean-centered at 0 (almost always done) before downstream analysis, this constant baseline difference cancels out completely. After standardization, the REF-scored and ALT-scored models produce perfectly identical values for every individual.

Allele Labels are Relative Constructs

What does it mean for an allele to be REF or ALT? Major or Minor?

• REF and ALT depend entirely on the specific reference genome assembly used for alignment (e.g., GRCh37 vs. GRCh38). An ALT allele in one assembly might be the REF allele in another. If the person used as a reference has a rare variant, the the REF allele is ALT for almost everyone in the world.
• Designating things as Major and Minor is equally problematic. Now, depending entirely on the allele frequencies within a specific study cohort or subpopulation, a "C" allele might be minor in one population but major in another population. E.g. the minor allele in a European cohort could be the major allele in Africa or East Asia.

Since these designations therefore mean very little biologically (technical or study-relative, but not deep biology) it makes sense that relying strictly on REF/ALT designations would flip an effect direction but in reality change nothing when comparing scores across different studies. When applying weights, it is necessary to align the effect sizes to the actual nucleotide sequences (A, C, T, G) rather than their relative labels.