Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces PerturbEnergy, a novel conditional energy-based prior integrated into autoencoder architectures for cellular perturbation prediction, to disentangle perturbation-specific signals from cellular complexity. By assigning lower energy to perturbation-compatible latent representations, it explicitly biases the latent space towards perturbation-preferred regions, demonstrably improving perturbation recognition and achieving state-of-the-art DEG prediction performance. The method's effectiveness is validated across four benchmark datasets, showing significant gains in perturbation recognition and competitive results in gene response prediction.
A conditional energy prior explicitly guides latent space towards perturbation-specific signals, unlocking an 83.2% improvement in perturbation recognition for cellular response prediction.
AI-based cellular perturbation prediction is essential to accelerate drug discovery and therapeutic identification. However, perturbation-specific signals are generally obscured by cellular complexity, making it difficult for existing methods to correctly capture perturbation effects. To address this problem, we propose PerturbEnergy, which introduces a conditional energy-based prior over perturbation representations. It assigns lower energy (higher probability) to representations compatible with the given perturbation and higher energy to others, thereby explicitly tilting the latent space toward perturbation-preferred regions. Besides, PerturbEnergy can be integrated into autoencoder architectures as we find its Evidence Lower Bound (ELBO) is equivalent to the ELBO of the complete autoencoder. We further prove that optimizing PerturbEnergy increases the lower bound on the conditional mutual information between representation and perturbation. Extensive experiments on four benchmark datasets demonstrate the effectiveness of our PerturbEnergy: it improves perturbation recognition especially in latent space with improvement of up to 83.2\%; It also achieves state-of-the-art performance for predicting top-50/100/200 differentially expressed gene(DEG) response and remaining competitive on highly variable gene(HVG) prediction. The code is provided: https://github.com/BlueWhaleLab/PerturbEnergy
NUS