The flawed assumption
Publicly available algorithms failed to detect the key mutation type. Investment teams mistakenly concluded WRN was an attractive target because the invalidating evidence was "absent."
Investment lessons · WRN
The $400M algorithm mistake
More than $400M went into WRN inhibitors, with minimal benefit for patients. Gordion’s analysis of patient data found a type of WRN mutation that commonly used algorithms miss, showing that WRN is not essential for cancer cell survival. Absence of evidence is not evidence of absence.
The WRN helicase · by Pawel Zawadzki, PhD
$400M+
Invested by multiple companies
The case
A costly failure
Despite massive investment from multiple companies, clinical trials for WRN inhibitors have repeatedly shown minimal benefits for cancer patients.
This outcome highlights a stark disconnect between strong data from preclinical models and the complex reality of patient responses.
Patient data
What the patient data revealed
Deep, AI-augmented analysis of patient data uncovered an error produced by a commonly used algorithm:
- A previously undetected type of WRN mutation was identified by Gordion technology.
- This mutation proved that WRN is not essential for cancer cell survival, hence a poor drug target. Actually, it appears WRN is rather a tumor suppressor!
- This single, patient data-derived finding directly explained the widespread failure of the clinical trials of WRN inhibitors.
Mutations inactivating both copies of a gene
The take-home message
Absence of evidence is not evidence of absence
AI warning
This is a critical lesson for all AI users: the origins of AI-generated insights can be unclear. However, even a greater challenge is assessing what your models might be overlooking.
Recommendations
Consult patient data for reliable outcomes
01
Be skeptical about computational tools, as each has its own limitations.
A proper AI stack can eliminate these problems.
02
Target selection is still oncology’s biggest single-point failure.
Patient data and well-designed AI are the only reliable way to know which targets truly matter.
03
But do not delay AI adoption, as others will outcompete you.
Select your AI partner wisely, as this will define your future.
Medicines intended to work in patients should be developed with patient data in mind.