Efficient Data Selection for Training Genomic Perturbation Models
Not on the map. Why?
Abstract (arXiv)
Genomic studies face a vast hypothesis space, while interventions such as gene perturbations remain costly and time-consuming. To accelerate such experiments, gene perturbation models predict the transcriptional outcome of interventions. Since constructing the training set is challenging, active learning is often employed in a "lab-in-the-loop" process. While this strategy makes training more targeted, it is substantially slower, as it fails to exploit the inherent parallelizability of Perturb-seq experiments. Here, we focus on graph neural network-based gene perturbation models and propose a subset selection method that, unlike active learning, selects the training perturbations in one shot. Our method chooses the interventions that maximize the propagation of the supervision signal to the model. The selection criterion is defined over the input knowledge graph and is optimized with submodular maximization, ensuring a near-optimal guarantee. Experimental results across multiple datasets show that, in addition to providing months of acceleration compared to active learning, the method improves the stability of perturbation choices while maintaining competitive predictive accuracy.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 28 of its 56 references that are in the map. First posted on arXiv 18 Mar 2025.
Builds on
28 of 56 references.
References · 28
- 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,361
- 2019Simplifying Graph Convolutional Networkscited by 1,477
- 2019Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Boundscited by 610
- 2019BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learningcited by 432
- 2022Active Learning on a Budget: Opposite Strategies Suit High and Low Budgetscited by 133
- 2020Unifying Graph Convolutional Neural Networks and Label Propagationcited by 98
- 2021Shift-Robust GNNs: Overcoming the Limitations of Localized Graph Training Datacited by 87
- 2021Subgroup Generalization and Fairness of Graph Neural Networkscited by 69
- 2019Bayesian Batch Active Learning as Sparse Subset Approximationcited by 68
- 2022Interventions, Where and How? Experimental Design for Causal Models at Scalecited by 49
- 2022A Framework and Benchmark for Deep Batch Active Learning for Regressioncited by 49
- 2023Towards Understanding the Generalization of Graph Neural Networkscited by 41
16 more
- 2023Generalization in Graph Neural Networks: Improved PAC-Bayesian Bounds on Graph Diffusioncited by 40
- 2019Rethinking deep active learning: Using unlabeled data at model trainingcited by 39
- 2019Parting with Illusions about Deep Active Learningcited by 38
- 2023Modelling Cellular Perturbations with the Sparse Additive Mechanism Shift Variational Autoencodercited by 38
- 2022Learning Causal Representations of Single Cells via Sparse Mechanism Shift Modelingcited by 34
- 2018Addressing the Item Cold-start Problem by Attribute-driven Active Learningcited by 33
- 2021Grain: Improving Data Efficiency of Graph Neural Networks via Diversified Influence Maximizationcited by 28
- 2021RIM: Reliable Influence-based Active Learning on Graphscited by 23
- 2021GeneDisco: A Benchmark for Experimental Design in Drug Discoverycited by 18
- 2021Near-Optimal Multi-Perturbation Experimental Design for Causal Structure Learningcited by 17
- 2022Information Gain Propagation: a new way to Graph Active Learning with Soft Labelscited by 16
- 2023DiscoBAX: Discovery of Optimal Intervention Sets in Genomic Experiment Designcited by 12
- 2024Uncertainty for Active Learning on Graphscited by 12
- 2024Season combinatorial intervention predictions with Salt & Pepercited by 11
- 2022Neural Design for Genetic Perturbation Experimentscited by 6
- 2021Low-Regret Active learningcited by 3
Built on by
Not cited yet.