What question did this study set out to answer?

To systematically evaluate the capability of diverse DNA foundation models to predict the functional consequences of genetic variants using zero-shot scoring across local and extended contexts.

August 19, 2026Genome biologyOpen Access

Benchmarking DNA foundation models for zero-shot variant effect prediction shows the importance of context, training, and architecture

View Full Paper

Ask AI

Bookmark

Authors

IAIlaria Alfisi FCFrancesca Ciapi MBMarta Baragli

Discussion

Member takes

Overview

Benchmarking evaluation shows multi-species DNA models excel at zero-shot variant effect prediction, indicating that training data diversity and model scale govern genomic accuracy.

Key Points

To systematically evaluate the capability of diverse DNA foundation models to predict the functional consequences of genetic variants using zero-shot scoring across local and extended contexts.
Benchmarked four DNA foundation model families (NT, DNABERT, HyenaDNA, and Evo 2) without task-specific fine-tuning.
Assessed sensitivity to sequence alterations using pathogenic, benign, and uncertain single nucleotide variants (SNVs) sourced from ClinVar across target and adjacent sequence tokens.
Large multi-species NT models outperformed other architectures in classifying variant pathogenicity, demonstrating superior discriminative power when aggregating zero-shot scores over surrounding tokens.
Models trained exclusively on human sequences or optimized via next-token prediction (DNABERT, HyenaDNA, and Evo 2) exhibited restricted contextual awareness and reduced variant discrimination.

Cite This Study

Alfisi et al. (2026) studied this question.

synapsesocial.com/papers/6a85643903308d306e2d7e41 https://doi.org/10.1186/s13059-026-04238-0

View Full Paper

Ask AI

Bookmark

August 19, 2026Genome biologyOpen Access

Benchmarking DNA foundation models for zero-shot variant effect prediction shows the importance of context, training, and architecture

View Full Paper

Ask AI

Bookmark

Authors

IAIlaria Alfisi FCFrancesca Ciapi MBMarta Baragli

Discussion

Member takes

Overview

Benchmarking evaluation shows multi-species DNA models excel at zero-shot variant effect prediction, indicating that training data diversity and model scale govern genomic accuracy.

Key Points

To systematically evaluate the capability of diverse DNA foundation models to predict the functional consequences of genetic variants using zero-shot scoring across local and extended contexts.
Benchmarked four DNA foundation model families (NT, DNABERT, HyenaDNA, and Evo 2) without task-specific fine-tuning.
Assessed sensitivity to sequence alterations using pathogenic, benign, and uncertain single nucleotide variants (SNVs) sourced from ClinVar across target and adjacent sequence tokens.
Large multi-species NT models outperformed other architectures in classifying variant pathogenicity, demonstrating superior discriminative power when aggregating zero-shot scores over surrounding tokens.
Models trained exclusively on human sequences or optimized via next-token prediction (DNABERT, HyenaDNA, and Evo 2) exhibited restricted contextual awareness and reduced variant discrimination.

Cite This Study

Alfisi et al. (2026) studied this question.

synapsesocial.com/papers/6a85643903308d306e2d7e41 https://doi.org/10.1186/s13059-026-04238-0

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark