13. Exhaustive screening — what MObyDiCK does · Dr. Seok's Cancer-Cell Digital Twin, First Steps

Last lesson we chose DDX54 and switched it off. But why DDX54? If you picked it already knowing the answer, that is not a prediction.

Virtual screening takes the opposite line.

Do not choose. Try everything.
Then rank the results and let the computation choose the answer.

That is the skeleton of what a platform like Prof. Cho's MObyDiCK does. At a bench one gene takes weeks; here the whole lot goes by in a blink.

This takes well under a second in your browser.

Python Runs in your browser

import bn

net = bn.twin()

# Candidate targets - the 'gauge' nodes we read the phenotype off are excluded (why: lesson 16)
print("candidate targets:", net.druggable())
print("excluded nodes   :", list(net.readouts), "  <- the gauges we read the result from")
print()

n_targets = len(net.druggable())
print(f"single perturbations = {n_targets} targets x (off, on) 2 = {n_targets * 2} runs")
print(f"each run rolls all {2 ** net.n} initial states to the end, so")
print(f"{n_targets * 2 * 2 ** net.n:,} cell simulations in total")

reversion = the share of once-cancerous cells that returned to normal. harm = how many normal ones became cancer.

Python Runs in your browser

import bn

net = bn.twin()
net.screen()          # try every single perturbation and print the scorecard

It splits three ways. Remember that PDL1 and CD47 are in the middle group (no effect).

Python Runs in your browser

import bn

net = bn.twin()
rows = net.screen(quiet=True)

# Split the treatments into successes and failures
win  = [r for r in rows if r["reversion"] == 100 and r["harm"] == 0]
fail = [r for r in rows if r["reversion"] == 0 and r["harm"] == 0 and r["treatment"] != "untreated"]
bad  = [r for r in rows if r["harm"] > 0]

print("treatments that reverted it completely")
for r in win:
    print("   [ok]", r["treatment"])
print("\ntreatments with no effect at all")
for r in fail:
    print("   [--]", r["treatment"])
print("\ntreatments that actively harm")
for r in bad:
    print(f"   [X]  {r['treatment']}   turned {r['harm']} normal states into cancer")

🎮 Screening table

Press a column heading to sort. What you rank by is what changes the 'best target'.

Reversion, harm, cancer left — the winner changes with the column you sort on.

Loading…

🤔 Check yourself

Why does virtual screening 'try everything' instead of choosing targets first?

🤔 Check yourself

How should you treat a screening row whose 'harm' column is not zero?

🤔 Check yourself

Reading it off the circuit, why did PDL1 knockout alone do nothing here?

Answer T_Kill* = Proliferation and not PDL1 and not CD47. There are two brakes; release one and the car stays exactly where it was.