88.9%
share of Mosaic's 9 tested designs that bound the Nipah virus target, against 9.6% of all 1,026 designs
What the chart shows
Adaptyv Bio’s 2025 challenge asked teams to design proteins that bind a Nipah virus protein, then tested the designs in the laboratory. Of 1,026 designs tested, 99 (9.6%) bound the target, and the AI Index reports that none neutralised it.
Mosaic, a specialised method that combines several tools with expert tuning, had the highest rate: 8 of its 9 designs (88.9%) bound. RFdiffusion reached 17.6%, BindCraft 7.4% and BoltzGen 1.9% of 107 designs; none of Germinal’s 14 bound. Most designs could be produced: 85.9% expressed.
Key numbers
| Measure | Value |
|---|---|
| All methods, bound (1,026 designs) | 9.65% |
| Mosaic, bound (9) | 88.89% |
| RFdiffusion, bound (17) | 17.65% |
| BindCraft, bound (27) | 7.41% |
| BoltzGen, bound (107) | 1.87% |
| All methods, expressed | 85.87% |
How to read it
- Bars show binding rates, the second block of the data file; the first block gives expression rates, the share of designs the laboratory could produce.
- Design counts per method are from the AI Index’s chart labels; the five methods shown account for 174 of the 1,026 designs.
- Samples are small: one design more or fewer changes Mosaic’s rate by 11 percentage points.
Source: Adaptyv Nipah Binder challenge, 2026, via the AI Index Report 2026 (Stanford HAI), figure 6.1.10: Protein design success rates in Adaptyv Nipah Binder challenge. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion