Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines | Nature Methods
The analysis presented in this Brief Communication shows that, despite their complexity, current deep learning models do not outperform linear baselines in predicting gene perturbation effects, thus emphasizing the importance of further method development and thorough evaluation.
Brief Communication Open access Published: 04 August 2025 Constantin Ahlmann-Eltze ORCID: orcid.org/0000-0002-3762-068X1,2 nAff3, Wolfgang Huber ORCID: orcid.org/0000-0002-0474-22182 & Simon Anders ORCID: orcid.org/0000-0003-4868-18051 Nature Methods volume 22, pages 1657–1661 (2025) Cite this article 118k Accesses 250 Citations 200 Altmetric Metrics details Abstract Recent research in deep-learning-based foundation models promises to learn representations of single-cell data that enable prediction of the effects of genetic perturbations. Here we compared five foundation models…
related reading
- Evaluating the role of pretraining dataset size and diversity on single-cell foundation model performancenature.com
- Advancing regulatory variant effect prediction with AlphaGenomenature.com
- PREDICTING SINGLE-CELL PERTURBATION RESPONSES FOR UNSEEN DRUGSarxiv.org
- Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkersgoodfire.ai
- scGen predicts single-cell perturbation responsesnature.com
- Accurate predictions on small data with a tabular foundation modelnature.com
- Quantifying the effect of experimental perturbations at single-cell resolutionnature.com
- On the Opportunities and Risks of Foundation Modelsarxiv.org
- Predicting cellular responses to complex perturbations in high‐throughput screens | Molecular Systems Biology | Springer Nature Linkembopress.org
- Towards the explainability of protein language modelsnature.com
- Leveraging supervised learning for functionally informed fine-mapping of cis-eQTLs identifies an additional 20,913 putative causal eQTLsnature.com
- 1810.04805arxiv.org