MELD: Mutation Effect Learning through Differential embeddings

Abstract

We introduce MELD (Mutation Effect Learning through Differential Embeddings), a compact model for predicting the functional consequences of single nucleotide variants (SNVs) by training directly on the wild-type (WT) versus mutant differential. Interpreting SNVs is central to clinical genomics, yet experimental characterization cannot scale to millions of observed variants. Leading sequence-to-function models (Enformer, Borzoi, AlphaGenome) are trained on absolute genomic tracks and score variants post hoc through in-silico mutagenesis (ISM), so the WT-mutant difference never enters the training loss. MELD instead places the logit difference {Delta}z = zmut - zwt inside the supervised objective. It combines a shared CNN-Transformer encoder, a mutant-specific mutation-mask boost, multiscale mutation-centered pooling, and a learned adaptive amplification factor . On ClinVar (1.57M variants), MELD (1.9M parameters) reached an AUROC of 0.986 versus 0.800 for a same-data, same-backbone ISM baseline, with ablations confirming the contribution of each component. On a shared benchmark, it outperformed frozen foundation models with 3-1,300x more parameters (AUROC 0.967 vs. 0.924 for AlphaGenome and 0.936 for Nucleotide Transformer). With only 50, 000 training variants, it exceeded the ISM baseline trained on all 1.48M (AUROC 0.903 vs. 0.796) variants. On BRCA1 saturation genome editing, frozen MELD achieved a Spearman {rho} of 0.42, above Borzoi (0.40), and ClinVar transfer raised it to 0.45. These results demonstrate that direct differential supervision is a compact, data-efficient alternative to post-hoc ISM scoring with substantially larger pretrained models.

Representation learning of somatic missense mutation landscapes across cancer types
Z.-L. Wang, M. Cai, Z. Li +7
bioRxiv25 Oct 2026PreprintLicence not statedFull textBioinformatics & computational biologybioRxiv
Probing Physical Intuition in Large Language Models through Partner-Dependent Mutation Prediction
L. Cheng, K. Zheng, W. Huang +2
bioRxiv9 Oct 2026PreprintCC BY-NC-NDFull textBioinformatics & computational biologybioRxiv
Mutation in Ralstonia pseudosolanacearum F1C1 cytochrome p450 homologue impairs virulence through root but not through shoot of tomato plants
L. Dutta, K. Kabyashree, S. J. Giri +11
bioRxiv9 Oct 2026PreprintCC BYFull textPlant biologybioRxiv
NeuroLens: Learning Latent Embeddings of Neural Semantics from Chronic Recordings
Hanrui Lyu, Baiyuan Chen, Tianshu Tan +8
arXiv2 Oct 2026PreprintarXiv licenceFull textNeurosciencearXiv
Conserved synaptic modifications by humanized tau through diverse isoform- and mutation-specific intracellular interactions
E. Nickerson, J. Pomeroy-Tuck, S. Pratt +6
bioRxiv3 Nov 2026PreprintCC BY-NDFull textNeurosciencebioRxiv

Cite

BibTeX RIS

BibTeX

@misc{alajji2026meld,
  title = {{MELD: Mutation Effect Learning through Differential embeddings}},
  author = {A. Alajji and R. Jashil and R. Mall and M. Elshrif},
  howpublished = {bioRxiv},
  year = {2026},
  month = oct,
  doi = {10.64898/2026.10.29.762727},
  url = {https://www.biorxiv.org/content/10.64898/2026.10.29.762727v1},
  note = {CC BY-NC-ND},
}

RIS

TY  - UNPB
TI  - MELD: Mutation Effect Learning through Differential embeddings
AU  - A. Alajji
AU  - R. Jashil
AU  - R. Mall
AU  - M. Elshrif
PY  - 2026
DA  - 2026/10/29
PB  - bioRxiv
DO  - 10.64898/2026.10.29.762727
UR  - https://www.biorxiv.org/content/10.64898/2026.10.29.762727v1
AB  - We introduce MELD (Mutation Effect Learning through Differential Embeddings), a compact model for predicting the functional consequences of single nucleotide variants (SNVs) by training directly on the wild-type (WT) versus mutant differential. Interpreting SNVs is central to clinical genomics, yet experimental characterization cannot scale to millions of observed variants. Leading sequence-to-function models (Enformer, Borzoi, AlphaGenome) are trained on absolute genomic tracks and score variants post hoc through in-silico mutagenesis (ISM), so the WT-mutant difference never enters the training loss. MELD instead places the logit difference {Delta}z = zmut - zwt inside the supervised objective. It combines a shared CNN-Transformer encoder, a mutant-specific mutation-mask boost, multiscale mutation-centered pooling, and a learned adaptive amplification factor . On ClinVar (1.57M variants), MELD (1.9M parameters) reached an AUROC of 0.986 versus 0.800 for a same-data, same-backbone ISM baseline, with ablations confirming the contribution of each component. On a shared benchmark, it outperformed frozen foundation models with 3-1,300x more parameters (AUROC 0.967 vs. 0.924 for AlphaGenome and 0.936 for Nucleotide Transformer). With only 50, 000 training variants, it exceeded the ISM baseline trained on all 1.48M (AUROC 0.903 vs. 0.796) variants. On BRCA1 saturation genome editing, frozen MELD achieved a Spearman {rho} of 0.42, above Borzoi (0.40), and ClinVar transfer raised it to 0.45. These results demonstrate that direct differential supervision is a compact, data-efficient alternative to post-hoc ISM scoring with substantially larger pretrained models.
ER  -

Abstract

Related

Cite