6. Naming things — which attractor is 'cancer'? · Dr. Seok's Cancer-Cell Digital Twin, First Steps

Everything the computer did up to the last lesson amounts to this — it found four attractors. That is all.

Calling one of them "the malignant cancer state" was us. The grounds were that a gene named Cancer_Growth is 1, and giving that gene that name was also us.

This is not a weakness; it is a premise that must be stated out loud. A number like 'cancer 25%' that comes out of a model depends entirely on the naming rule.

Four attractors, and a single digit as the criterion.

Python Runs in your browser

import bn

net = bn.mini()
land = net.landscape()

# See for yourself how the label gets attached to each attractor
for a, basin in sorted(land.basins.items(), key=lambda kv: -len(kv[1])):
    d = dict(zip(net.genes, a[0]))
    print(f"{net.label(a):<10} <- Cancer_Growth={d['Cancer_Growth']}   "
          f"basin of {len(basin):>2}   ON: {', '.join(net.on_genes(a[0])) or '(none)'}")

print()
print("Exactly one digit, Cancer_Growth, did all the judging.")

This is why quoting a number obliges you to say by what criterion it was counted.

Python Runs in your browser

import bn

net = bn.mini()

# Change the calling rule and the 'cancer fraction' changes wholesale
def count_with(rule):
    land = net.landscape()
    return sum(1 for a, b in land.basins.items()
               if rule(dict(zip(net.genes, a[0]))) for _ in b)

rules = {
    "Cancer_Growth = 1 (our rule)":       lambda d: d["Cancer_Growth"] == 1,
    "CD47 = 1 (immune evasion only)":     lambda d: d["CD47"] == 1,
    "p53 = 0 (broken brake only)":        lambda d: d["p53"] == 0,
    "CD47 = 1 and p53 = 0":               lambda d: d["CD47"] == 1 and d["p53"] == 0,
}
for name, r in rules.items():
    n = count_with(r)
    print(bn.pad(name, 38) + f"counted as cancer {n:>2}/16  ({n/16*100:5.1f}%)")

print()
print("Same computation, same attractors. The only thing that changed is the naming rule.")

🎮 Name-tag bench

Change the criterion and watch which cells get counted as 'cancer'.

Pick a criterion and all 16 cells are repainted.

Loading…

🤔 Check yourself

A model reports 'cancer states: 25%'. What should you check first?

🤔 Check yourself

What is the biggest benefit of writing the phenotype rule down as code?

🤔 Check yourself

What is the difference between 'finding an attractor' and 'naming an attractor'?