Leveraging genomic deep learning models for the prediction of noncoding variant effects.

Genomic deep learning models can predict noncoding variant effects, with supervised sequence-to-activity and self-supervised language models showing promise.

• Why it matters: Understanding noncoding variant function is a major challenge in human genetics, and accurate predictions are needed to interpret genetic variation's role in health and disease.
• What they did: The review discusses progress in using these models for variant effect prediction, practical prediction considerations, and evaluation methods based on ground-truth data.
• The result: Findings highlight current model strengths and limitations, guiding practitioners and identifying opportunities to improve model development and assessment.