SCALE:Scalable Conditional Atlas-Level Endpoint transport for virtual cell perturbation prediction

2026-03-18

AI Summary

This work addresses key limitations in current virtual cell perturbation prediction methods—namely, inefficient training and inference pipelines, modeling instability under high-dimensional sparse representations, and evaluation metrics overly reliant on reconstruction accuracy at the expense of biological fidelity. To overcome these challenges, we propose a perturbation modeling framework grounded in conditional optimal transport, integrating a set-aware flow architecture with endpoint-guided supervision to jointly enhance training stability and perturbation effect recovery. Leveraging an efficient BioNeMo-based training-inference system and an LLaMA-driven cellular encoder, our approach achieves substantial gains in scalability. On the Tahoe-100M benchmark, it outperforms the STATE model by 12.02% in PDCorr and 10.66% in DE Overlap, while accelerating pretraining by 12.51× and inference by 1.29×.

Technology Category

• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Virtual cell models aim to enable in silico experimentation by predicting how cells respond to genetic, chemical, or cytokine perturbations from single-cell measurements. In practice, however, large-scale perturbation prediction remains constrained by three coupled bottlenecks: inefficient training and inference pipelines, unstable modeling in high-dimensional sparse expression space, and evaluation protocols that overemphasize reconstruction-like accuracy while underestimating biological fidelity. In this work we present a specialized large-scale foundation model SCALE for virtual cell perturbation prediction that addresses the above limitations jointly. First, we build a BioNeMo-based training and inference framework that substantially improves data throughput, distributed scalability, and deployment efficiency, yielding 12.51* speedup on pretrain and 1.29* on inference over the prior SOTA pipeline under matched system settings. Second, we formulate perturbation prediction as conditional transport and implement it with a set-aware flow architecture that couples LLaMA-based cellular encoding with endpoint-oriented supervision. This design yields more stable training and stronger recovery of perturbation effects. Third, we evaluate the model on Tahoe-100M using a rigorous cell-level protocol centered on biologically meaningful metrics rather than reconstruction alone. On this benchmark, our model improves PDCorr by 12.02% and DE Overlap by 10.66% over STATE. Together, these results suggest that advancing virtual cells requires not only better generative objectives, but also the co-design of scalable infrastructure, stable transport modeling, and biologically faithful evaluation.

Problem

• virtual cell perturbation prediction
• scalable training and inference
• high-dimensional sparse expression
• biological fidelity evaluation
• conditional transport

Innovation

• conditional transport
• scalable foundation model
• virtual cell perturbation
• set-aware flow architecture
• biologically faithful evaluation