3billion Named Top-Performing Team at CAGI7, the Global Benchmark for AI Genome Interpretation

3billion announced that its AI research team was selected as the Top-Performing Team in the FGFR Variant Functional Prediction Challenge at CAGI7 (Critical Assessment of Genome Interpretation), held December 6–8 in Boston.

CAGI, now in its 7th year, is an internationally recognized competition supported by the U.S. National Institutes of Health (NIH) and organized by UC Berkeley. The event evaluates how accurately AI models can predict the functional impact of genetic variants using a fully blinded assessment. It is regarded as the world’s leading benchmark for comparing genome interpretation technologies across academia and industry.

3billion previously distinguished itself at CAGI6 (2022), where its models outperformed state-of-the-art AI systems, including Google DeepMind’s AlphaMissense, in whole-genome rare disease diagnosis and protein function prediction tasks.

Breakthrough Performance in FGFR Variant Interpretation

In CAGI7, 3billion competed in the FGFR (Fibroblast Growth Factor Receptor) functional prediction challenge, a high-complexity task requiring models to estimate how every possible missense substitution across FGFR1–4 affects protein function.

Unlike disease-specific prediction tasks, this challenge focuses on a model’s ability to accurately distinguish how individual variants shift functional outcomes—an ability essential to both clinical variant interpretation and precision drug development.

To address this, 3billion developed ClearVariantPro (CVP), a multi-task learning AI model optimized through a target-specific embedding space. The model was evaluated alongside leading AI systems such as AlphaMissense and Meta’s ESM.

While many existing models showed performance bias—accurate in predicting either Gain-of-Function (GoF) or Loss-of-Function (LoF) but not both—CVP was the only model among all participants to demonstrate high precision in both functional directions, leading to its selection as the top performer.

The evaluation was conducted by a research team led by Professor Russ Altman, a globally recognized expert in pharmacogenomics at Stanford University. CVP’s strong performance outpaced competing models across metrics, with particularly large margins in GoF prediction accuracy.

Validation Across Diagnostics and Drug Discovery Applications

The results confirm that 3billion’s AI technology possesses world-class capability in finely resolving functional effects of genetic variants—a foundation critical for:

• advancing rare disease diagnosis,
• accurately identifying pathogenic mechanisms,
• discovering variant-driven therapeutic targets, and
• predicting protein behavior for precision medicine and drug development.

At CAGI7, 3billion achieved top-tier results across three major challenges, further validating the scalability and scientific robustness of its genome interpretation platform.

CEO Statement

3billion CEO Changwon Keum stated:
“CAGI provides a globally unified standard for evaluating genome interpretation technologies under identical conditions. Being named the top-performing team at CAGI7 reaffirms the global leadership of our variant interpretation technology. We will continue expanding our applications—from rare disease diagnostics to precision medicine and drug discovery.”

Several CAGI7 challenges will continue evaluation beyond the conference, with full reports and associated publications expected to be released throughout next year.