Deep learning models outperform linear regression by capturing epistatic variance

Association·arcadia

Claim that deep learning models outperform linear regression in genotype-phenotype prediction tasks by capturing epistatic interactions that linear models cannot.

Confidence

90%

active

Evidence Quote

“DL models can capture epistatic variance under conditions where epistasis is statistically apparent, outperforming linear regression in G→P prediction.”

Relationship

Deep learning methods outperforms Epistatic interactions

Arguments

Deep learning methodssubject

Epistatic interactionsobject

Connections (7)

Deep learning capture of epistatic variance in genotype-phenotype mappingInferenceChain

Non-additive interactions like epistasis drive phenotypic variationAssociation

Data regimes and deep learning benefits in genotype-phenotype mappingInferenceChain

Data regimes and genetic architecture constrain deep learning benefitsInferenceChain

Reasoning on deep learning benchmarking in genotype–phenotype mappingInferenceChain

Deep learning architecture and theoretical foundationsInferenceChain

Reasoning on deep learning and genotype-phenotype predictionInferenceChain

Evidence

“All code, including analysis notebooks, synthetic phenotype generator, and autoencoder model related to the study, made available on Zenodo.”

Author Unknown (2024). When do deep learning models outperform linear regression? doi:10.5281/zenodo.15644565 ↗

“Simulated data and analysis scripts made publicly available on Zenodo for reproducibility of genotype-phenotype mapping results.”

Synthesis for current paper

“Evidence line describing benchmarks performed with different architectures for predicting gene expression from DNA sequence reported by Barbadilla-Martínez et al. (2025).”

Barbadilla-Martínez et al. 2025

“Compiled empirical results from various deep learning benchmarking studies showing heterogeneity in model performance across datasets and genetic architectures.”

Synthesis for current paper

Association·arcadia