5. 5,202 slides — a different world per cancer type · Dr. Seok's Tumour-Infiltrating Lymphocytes, First Steps

Lessons 1–4 turned one map into one number. The table of that calculation run over all 5,202 slides of 13 cancer types that TCIA released ships inside the browser. (It is exactly the same code — til.metrics() called 5,202 times.)

Slide name, cancer type, infiltration rate, tissue tiles, cluster count, depth, clinical data, survival.

Python Runs in your browser

import til

d = til.slides()
print("table shape:", d.shape)
print()
print(d[["slide", "tumor", "til_pct", "tissue_patches", "n_clusters", "OS_time"]].head(5).to_string(index=False))
print()
print("columns:", ", ".join(d.columns))

The median runs from 0.00% (uveal melanoma) up to 12.28% (stomach).

Python Runs in your browser

import til

d = til.slides()
g = d.groupby("tumor").til_pct
t = g.describe()[["count", "25%", "50%", "75%", "max"]].round(2)
t.columns = ["slides", "Q1", "median", "Q3", "max"]
t["cancer"] = [til.NAMES_EN[x] for x in t.index]
print(t.sort_values("median", ascending=False).to_string())

Look at UVM (uveal melanoma) at the bottom. The median is 0.00%. More than half of its 63 slides have not a single red cell.

That is not a bug. The uvea inside the eye is an immune-privileged site — evolution walled it off so immune cells cannot barge in, because inflammation in the eye costs you your sight. That fact is printed straight onto the map.

At the other end, STAD (stomach), READ (rectum) and CESC (cervix) have medians of 9–12%. These are organs whose mucosa always has immune cells on standby.

Click a cancer type to compare distributions. The peaks sit in quite different places.

Loading…

A box plot — the middle line is the median, the box is the interquartile range.

Python Runs in your browser

import til
import matplotlib.pyplot as plt

d = til.slides()
order = d.groupby("tumor").til_pct.median().sort_values().index

fig, ax = plt.subplots(figsize=(9, 4.4))
ax.boxplot([d[d.tumor == t].til_pct.clip(0, 45) for t in order], showfliers=False)
ax.set_xticks(range(1, len(order) + 1))
ax.set_xticklabels(list(order), rotation=45, fontsize=8)
ax.set_ylabel("TIL %")
ax.axhline(d.til_pct.median(), color="crimson", ls="--", lw=1)
plt.tight_layout()

print("One box is one cancer type. They overlap, but they clearly sit in different places.")

🤔 Check yourself

What is the most plausible reason the median TIL of uveal melanoma (UVM) is 0.00%?

🤔 Check yourself

What homework does 'the distribution differs a lot by cancer type' set for later lessons?