starplast.lipids

Membrane lipid composition: rows are lipid species, and the species are the parasite's own.

The third metabolite slot asks what the parasite's membranes are made of. A lipidome of an infected culture cannot answer it. The host cell is most of the lipid in the dish, and a lipid measured there is not thereby a parasite lipid -- which is why the slot sat empty while several Toxoplasma lipidomics papers were on the shelf. This source answers it because the vesicles were collected from tachyzoites AFTER egress, in host-cell-free medium, so what is in them the parasite carried out.

Why these read as parasite composition rather than host carry-over

The study grew the parasite in four host backgrounds -- fibroblast, intestinal epithelium, myotube, Vero -- whose own lipidomes are not remotely alike: between any two of them 1,018 to 1,362 species differ at FDR < 0.05. Between the vesicles the same parasite released in those four backgrounds, 0 to 4 do. A composition set by the host would move when the host is replaced; this one does not. Both numbers are in the shipped archive, and tests/test_lipids.py recomputes them rather than trusting the sentence -- the whole case for calling these parasite lipids rests on that contrast, so it is checked and not cited.

Two columns, because there are two questions

lipid_ev_level_log2 is how much of a species is present. It is the published processed matrix, averaged over the nineteen vesicle samples, and nothing else.

lipid_ev_vs_host_clr is how much more of it is present than in the host cell the parasite grew in. That question needs the loading difference removed first: vesicles carry far less total material than a cell, so every raw EV-minus-cell difference is negative (-7 to -14 log2) and says only that a vesicle is smaller than a cell. Centring each sample on its own mean before differencing asks about proportion instead of amount, which is the question the word "composition" is asking.

The published archive has its own EV-minus-cell column. It is not used: it correlates 0.92 with every reconstruction attempted here but matches none of them to better than 3 log units, so its transform is not known. A number whose definition cannot be reproduced cannot be verified, and this column is computed here from the published per-sample matrices instead, by the rule stated above.

The key, and why the name collision the metabolite table warns about does not arise

Rows join the metabolite table on the same normalised name. metabolites warns that matching a second study by compound NAME is lossy and is the first thing to fix when a second study arrives. This is that second study, and the cost is not paid here: lipid shorthand (PI 38:4, SM 34:1;O2) and polar metabolite common names (citrate, glutamine) are different naming systems that do not overlap, and none of the 194 species collides with any of the 1,102 compounds already in the table. That is asserted in the tests. A third study of polar metabolites will still need an identifier.