JEPA · 2026-11-03

IsoAlign: Isoform-Aware DNA-Protein Alignment via Joint-Embedding Predictive Architectures

Niousha Sadjadi, Arman Seyed-Ahmadi, Tony Tu, Micaela Elisa Consens, Kerr Ding, Alif Munim, Lila Kari, Zehui Li, Barry B. Rubin, Bo Wang, Vallijah Subasri

arXiv:2611.00241PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

IsoAlign is a multimodal genomic pretraining framework for isoform-aware DNA-protein representations using two JEPA architectures: DP-JEPA (predict protein representations from DNA via contrastive alignment) and IM-JEPA (mask protein isoforms and predict them using DNA anchors and visible isoform context). The authors evaluate DNA representations on tasks including fitness, variant effect prediction, gene expression, and TF binding.

Why it matters

The paper targets a limitation of DNA-only self-supervised objectives that treat genomic sequences as flat token streams and overlook alternative splicing. By grounding pretraining in transcript-to-protein-isoform alignment with JEPA objectives, the authors aim to learn representations that better reflect functional consequences of genomic variation and improve downstream performance.

Method

  • Dual-encoder framework aligning tokenized DNA and protein sequences into a shared embedding space with modality-specific encoders and projection heads; protein encoder is frozen by default.
  • DP-JEPA: predicts protein embeddings from DNA anchors in a shared latent space using contrastive alignment with stop-gradient targets per protein isoform slot.
  • IM-JEPA: models intra-gene isoform relationships by masking ~50% of isoforms and predicting masked protein isoforms using DNA anchors and remaining visible isoform context, with SIGReg anti-collapse regularization.

Limitation

The paper states that existing genomic foundation model pretraining strategies often fail to deliver consistent or meaningful performance gains over specialist models, and that naive DNA-only self-supervised pretraining lacks grounding necessary to map genomic variation to functional downstream consequences, often resulting in representations that lack clinical relevance and sensitivity to genomic heterogeneity.

Abstract (from arXiv)

Genomic foundation models (GFMs) have emerged as a powerful paradigm for learning DNA sequence representations, yet existing self-supervised pretraining strategies overlook alternative splicing and underperform compared to specialized models. To address this gap, we introduce IsoAlign, a multimodal framework that grounds genomic pretraining in isoform-aware DNA-protein alignment to learn biologically meaningful representations. IsoAlign aligns transcripts to their corresponding protein isoforms, capturing both coding sequences and broader regulatory contexts, using two Joint-Embedding Predictive Architectures (JEPA): DNA-Protein JEPA (DP-JEPA), which predicts protein representations from DNA using contrastive alignment, and Isoform-Masked JEPA (IM-JEPA), which models relationships among alternative isoforms of the same gene through masked protein prediction. We evaluate the learned DNA representations of these architectures across a multi-metric evaluation suite spanning tasks, including the prediction of pathogenicity, fitness, gene expression, and TF binding. Our results demonstrate that isoform-aware cross-modal alignment using JEPA significantly improves functional representations and downstream performance, establishing a promising direction for GFM pretraining.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-11-03 · Indexed by Pub-AI: 2026-11-03 UTC

preprint

IsoAlign: Isoform-Aware DNA-Protein Alignment via Joint-Embedding Predictive Architectures

Niousha Sadjadi, Arman Seyed-Ahmadi, Tony Tu, Micaela Elisa Consens, Kerr Ding, Alif Munim, Lila Kari, Zehui Li, Barry B. Rubin, Bo Wang, Vallijah Subasri

First announced by arXiv: 2026-11-03 · Original record ↗

arxiv
2611.00241
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-11-03 · First detected: 2026-11-03 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2611.00241v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

Genomic foundation models (GFMs) have emerged as a powerful paradigm for learning DNA sequence representations, yet existing self-supervised pretraining strategies overlook alternative splicing and underperform compared to specialized models. To address this gap, we introduce IsoAlign, a multimodal framework that grounds genomic pretraining in isoform-aware DNA-protein alignment to learn biologically meaningful representations. IsoAlign aligns transcripts to their corresponding protein isoforms, capturing both coding sequences and broader regulatory contexts, using two Joint-Embedding Predictive Architectures (JEPA): DNA-Protein JEPA (DP-JEPA), which predicts protein representations from DNA using contrastive alignment, and Isoform-Masked JEPA (IM-JEPA), which models relationships among alternative isoforms of the same gene through masked protein prediction. We evaluate the learned DNA representations of these architectures across a multi-metric evaluation suite spanning tasks, including the prediction of pathogenicity, fitness, gene expression, and TF binding. Our results demonstrate that isoform-aware cross-modal alignment using JEPA significantly improves functional representations and downstream performance, establishing a promising direction for GFM pretraining.