Accuracy, with the cases we get wrong

Every figure is computed from a raw count and a sample size, with a Wilson score interval. A point estimate without its interval and its n is a claim about a population made from a sample, with the sample hidden.

Foldamer

MechanismMetric Estimate95% CI nDataset
aggregation propensity catch rate 82.4% 75.0–88.0% n=131 CPAD amyloid-class validation (this project; systematic sample by length 6-10, n=131)
aggregation propensity miss rate 17.6% 12.0–25.0% n=131 CPAD amyloid-class validation (this project; systematic sample by length 6-10, n=131)
aggregation propensity sup35 subset of misses 56.5% 36.8–74.4% n=23 CPAD amyloid-class validation (this project) -- fraction of the 23 misses specifically from the Sup35/ERF-3 polar-zipper prion domain
aromatic pi stacking precision 71.1% 66.7–75.1% n=453 Yang, Yorke, Knowles & Buehler 2025, Science Advances, DOI 10.1126/sciadv.adv1971
aromatic pi stacking recall 56.9% 52.8–60.9% n=566 Yang, Yorke, Knowles & Buehler 2025, Science Advances, DOI 10.1126/sciadv.adv1971
aromatic pi stacking specificity 67.8% 63.1–72.2% n=407 Yang, Yorke, Knowles & Buehler 2025, Science Advances, DOI 10.1126/sciadv.adv1971
ionic self complementary precision 100.0% 70.1–100.0% n=9 Njirjak, Zuzic, Babic, Jankovic, Otovic, Kalafatovic & Mausa 2024, Nature Machine Intelligence, DOI 10.1038/s42256-024-00928-1
ionic self complementary recall 31.0% 17.3–49.2% n=29 Njirjak, Zuzic, Babic, Jankovic, Otovic, Kalafatovic & Mausa 2024, Nature Machine Intelligence, DOI 10.1038/s42256-024-00928-1
ionic self complementary specificity 100.0% 67.6–100.0% n=8 Njirjak, Zuzic, Babic, Jankovic, Otovic, Kalafatovic & Mausa 2024, Nature Machine Intelligence, DOI 10.1038/s42256-024-00928-1

Not externally benchmarked

These detectors contribute to verdicts and have no external benchmark. Their calibration field is null with a stated reason rather than an invented number. Listing only the measured ones would be advertising.

coiled-coil heptad repeat (Crick/COILS-class)

Calibrated against a small internal reference set, not yet externally benchmarked against a large labelled dataset, so there is no Wilson interval to publish.

No interval
amphipathic alpha-helix (Eisenberg hydrophobic-moment)

Calibrated against a small internal reference set, not yet externally benchmarked against a large labelled dataset, so there is no Wilson interval to publish.

No interval

Agents with nothing published yet

Antiagent

No benchmarks published yet.

Live
Target Watch

No benchmarks published yet.

Beta
Lapsis

No benchmarks published yet.

Building
Biomarker Readiness

No benchmarks published yet.

Building
Analysis Audit

No benchmarks published yet.

Planned