7. 【Lab】 One plate takes more than half · Prof. Kwon's TCR-HLA and CD4/CD8, First Steps

Last lesson we said "the skew of researchers". Now we measure it.

The question is simple. How many plates did these 181,411 rows come from?

HLA-A*02:01 alone is 57% — more than everything else put together.

Python Runs in your browser

# Class I — which plate takes how many rows
import thla

r = thla.restriction("I", top=10).rename(columns={"줄": "rows", "에피토프": "epitopes"})
print(r.to_string())
print()
print(f"the top one alone is {r['%'].iloc[0]}% of everything")
print(f"the top five are {r['%'].head(5).sum():.1f}%")

0.47%. That is the slice of plates our picture of T cell recognition was drawn on.

Python Runs in your browser

# How many plates appear here, and how many exist
import thla

F = thla.facts()
seen = F["n_allele_seen"]["I"]
known = 18559          # two-field HLA-A/B/C proteins registered in IMGT/HLA
print(f"class I alleles appearing in this table: {seen}")
print(f"HLA-A/B/C proteins registered in IMGT:   {known:,}")
print(f"-> {100*seen/known:.2f}%")
print()
print("class II:", F["n_allele_seen"]["II"], "alleles")

Four fragments are half of it. Remove one and the shape of the map changes.

Python Runs in your browser

# The fragment side is worse — how few fragments fill half the table
import thla

d = thla.vdj("I")
e = d.groupby("epitope").size().sort_values(ascending=False)
c = (e.cumsum() / e.sum())
print(f"distinct fragments {len(e):,}")
print(f"   top {int((c < 0.5).sum())+1} cover 50% of all rows")
print(f"   top {int((c < 0.9).sum())+1} cover 90% of all rows")
print()
for ep, n in e.head(6).items():
    print(f"   {ep:<12}{n:>7,} rows  {100*n/len(d):>5.1f}%")

Delete plates one at a time. It takes very few before the map collapses.

Loading…

But what share of people carry A*02:01?

Two numbers have to sit side by side here.

Share of the data

57%
of VDJdb class I records

Share of people

about 25%
of the world carrying this allele
cited from the literature · AFND

More than twice over-represented. And even 25% is only an average: it varies enormously between populations — high in people of European descent, far lower in sub-Saharan African and Indigenous American populations.

'Assuming independence' is flagged because HLA genes are actually inherited together (linkage disequilibrium).

Python Runs in your browser

# Using cited frequencies — what share of people do these plates cover
import thla

F = thla.facts()
print(F["cited_note_en"])
print()
for a, f in sorted(F["freq"].items(), key=lambda x: -x[1]):
    print(f"   {a:<14}{f:>5.1f}%")
print()
top3 = ["HLA-A*02:01", "HLA-A*01:01", "HLA-A*03:01"]
print(f"people carrying at least one of those three (assuming independence): {thla.covered(top3)}%")
print("-> the T cell recognition of everyone else is barely on this map.")

🤔 Check yourself

What share of VDJdb class I records is HLA-A*02:01?

🤔 Check yourself

What does this skew most accurately mean?

🤔 Check yourself

The 88 class I alleles in this table are what share of the proteins registered in IMGT?