0
0TxPert is a novel deep learning framework designed to predict how genetic changes alter a cell’s transcriptomic profile, even in scenarios where the specific changes have never been experimentally observed. The model distinguishes itself by integrating multiple knowledge graphs, such as STRING and Gene Ontology, to leverage existing biological data as a guide for its predictions. By utilizing graph neural networks, the system effectively simulates the impact of single or combined gene disruptions across various cell types. Research findings indicate that TxPert consistently outperforms existing computational methods and approaches the accuracy of experimental reproducibility in several benchmarks. The study also emphasizes the importance of using batch-matched controls and refined evaluation metrics to ensure that predictions reflect true biological signals rather than experimental noise. Ultimately, this tool aims to accelerate drug discovery by providing a reliable way to conduct virtual biological screens before moving to costly laboratory tests.
References:
Wenkel F, Tu W, Masschelein C, et al. TxPert: using multiple knowledge graphs for prediction of transcriptomic perturbation effects[J]. Nature Biotechnology, 2026: 1-8.
