Probing Genomic Foundation Models with Splice-Variant Perturbations
- 1712 North Inc.
Abstract
As benchmarks saturate, traditional validation methods fail to capture the probabilistic representations of genomic foundation models. This study probes self-supervised generalists (NTv3 650M, Evo2 40B, Genos 10B) alongside task-specific specialists (SpliceAI, AlphaGenome, Borzoi) using a mechanistically focused paradigm centered on expert-curated single-nucleotide substitutions within the splicing regions of the exon-rich OPA1 gene. Extending this analysis across the broader corpus of OPA1 mutations, variants of uncertain significance (VUS), and an independent 65-gene dataset of spliceogenic mutations demonstrates that generalists can match task-specific networks. Training data scale, genetic diversity, and evolutionary context are key for capturing RNA splicing syntax, whereas sparse routing may constrain performance. Mechanistically, foundation models exhibit a compute-asymmetry: the Evo2 40B model frontloads compute into massive parameter scale to resolve single-nucleotide variants within a narrow 425-bp context window, whereas the bidirectional NTv3 650M model and the Genos 10B mixture-of-experts (MoE) architecture backload compute to inference, requiring spatial logit aggregation and expansive context windows of up to 94 kb. Furthermore, classification accuracy across these foundation models benefitted from locus-specific thresholding rather than mutation-class calibration, exposing representation boundaries of the models.
Related
- Genomic foundation models do not yet boost common variant discovery beyond standard functional annotations
- Genomic splice-donor disruption rescues systemic disease in progeria
- Case Report: A novel deep intronic MFSD8 splice-altering variant causes CLN7 disease through pseudoexon inclusion
- EEF1D-Associated Neurodevelopmental Disorders: Evidence From Iranian Siblings With a Biallelic Splice Junction Loss Variant.
- Cockayne Syndrome Phenotype in a Sudanese Child With a Homozygous ERCC8 Splice-Site Variant of Uncertain Significance: A Case Report
BibTeX
@misc{alavi2026probing,
title = {{Probing Genomic Foundation Models with Splice-Variant Perturbations}},
author = {M. V. Alavi},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.02.756289},
url = {https://www.biorxiv.org/content/10.64898/2026.10.02.756289v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Probing Genomic Foundation Models with Splice-Variant Perturbations
AU - M. V. Alavi
PY - 2026
DA - 2026/10/02
PB - bioRxiv
DO - 10.64898/2026.10.02.756289
UR - https://www.biorxiv.org/content/10.64898/2026.10.02.756289v1
AB - As benchmarks saturate, traditional validation methods fail to capture the probabilistic representations of genomic foundation models. This study probes self-supervised generalists (NTv3 650M, Evo2 40B, Genos 10B) alongside task-specific specialists (SpliceAI, AlphaGenome, Borzoi) using a mechanistically focused paradigm centered on expert-curated single-nucleotide substitutions within the splicing regions of the exon-rich OPA1 gene. Extending this analysis across the broader corpus of OPA1 mutations, variants of uncertain significance (VUS), and an independent 65-gene dataset of spliceogenic mutations demonstrates that generalists can match task-specific networks. Training data scale, genetic diversity, and evolutionary context are key for capturing RNA splicing syntax, whereas sparse routing may constrain performance. Mechanistically, foundation models exhibit a compute-asymmetry: the Evo2 40B model frontloads compute into massive parameter scale to resolve single-nucleotide variants within a narrow 425-bp context window, whereas the bidirectional NTv3 650M model and the Genos 10B mixture-of-experts (MoE) architecture backload compute to inference, requiring spatial logit aggregation and expansive context windows of up to 94 kb. Furthermore, classification accuracy across these foundation models benefitted from locus-specific thresholding rather than mutation-class calibration, exposing representation boundaries of the models.
ER -