6. 【Lab】 Opening 181,411 rows — what is in here? · Prof. Kwon's TCR-HLA and CD4/CD8, First Steps

The first five lessons were pictures. From here on we count.

The data is VDJdb: a table collecting every published claim of the form "this TCR recognised this fragment". We use the 3 June 2026 release, and after keeping only human rows, 181,411 remain.

gene   cdr3               v         j          epitope     mhc_a         mhc_b   cls
TRB    CASSIRSSYEQYF      TRBV19    TRBJ2-7    GILGFVFTL   HLA-A*02:01   B2M     I

One row is one chain, not one cell. A TCR has an α chain and a β chain, and plenty of experiments read only one of them.

Class II is 3.5%. Not because the world is short of CD4 T cells.

Python Runs in your browser

# The skeleton of the table
import thla

d = thla.vdj()
print(f"rows {len(d):,}   columns {list(d.columns)}")
print()
print("chain      ", dict(d["gene"].value_counts()))
print("plate class", dict(d["cls"].value_counts()))
print()
print("class II share : "
      f"{100*(d['cls']=='II').mean():.1f}%")

CMV, EBV and influenza dominate — the ones easiest to find in healthy donor blood.

Python Runs in your browser

# Which pathogen did the fragment come from
import thla

d = thla.vdj()
v = d["ag_species"].value_counts().head(10)
for s, n in v.items():
    print(f"   {s:<18}{n:>7,}  " + "█" * max(1, int(60 * n / v.max())))
print()
print(f"HomoSapiens = self proteins and tumour antigens, {int(v.get('HomoSapiens', 0)):,} rows")

Keep only high scores and the table shrinks hard. Which to use depends on the question.

Python Runs in your browser

# Confidence score — VDJdb rates the strength of the evidence 0 to 3
import thla

d = thla.vdj()
mean_ = d["score"].mean()
for s in sorted(d["score"].unique()):
    n = int((d["score"] == s).sum())
    print(f"   score {s}  {n:>7,} rows  {100*n/len(d):>5.1f}%")
print()
print(f"mean {mean_:.2f} — nearly all of it is 0.")
print("0 does not mean 'wrong'. It means 'one source only'.")

Filter by class, chain and pathogen. You can see what disappears when you take something out.

Loading…

Why is class II only 3.5%?

The body has more CD4 T cells than CD8 — roughly 2:1 in peripheral blood. Yet class II is 3.5% of this table. Why?

🔴

Tetramers are hard

The MHC tetramers used to catch antigen-specific T cells are far easier to build for class I. Class II tetramers are awkward to make and awkward to use.

📏

The epitopes are long

Class II fragments run 9 to 24 residues with fuzzy edges. Deciding where the epitope starts and stops is genuinely difficult.

💉

The money went to CD8

Cancer immunotherapy's protagonist was CD8 for a long time. Funding followed.

One line: the 3.5% in this table is methodology, not biology.

🤔 Check yourself

What does a single row of this table represent?

🤔 Check yourself

What is the biggest reason class II records are only 3.5%?

🤔 Check yourself

What does a VDJdb score of 0 mean?