🎙️ Episode 64: Pisces — A Multi-Modal Framework for Predicting Drug Combination Synergy

🧬 In this episode of Base by Base, we delve into a groundbreaking methodological advance published by Xu et al. (2025) in Cell Genomics, which introduces Pisces, a novel machine-learning framework that overcomes data sparsity in drug combination studies by generating multiple modality-specific views for each drug pair. By projecting eight complementary data modalities—ranging from molecular structures and SMILES strings to drug targets, side-effect profiles, and ontological annotations—into a unified embedding space, Pisces expands the effective size of existing datasets up to sixty-four-fold, enabling robust synergy prediction even when some modalities are missing .

🔍 Highlights of the study: The Pisces framework creates independent augmented instances for each modality pair, bypassing the need for imputation; a tailored aggregator then selects the most informative predictions via a noisy-label ResNet module, excluding low-quality augmentations; across benchmarks, Pisces achieves state-of-the-art performance on cell-line-based and patient-derived xenograft synergy prediction tasks, significantly outperforming five leading approaches; it generalizes to never-before-seen drug combinations and cell lines, and even extends to predicting drug–drug interactions with high accuracy; interpretability analyses uncovered a breast cancer–sensitive pathway that correlates with improved survival in TCGA patients; finally, all code and datasets are openly accessible, accelerating future discoveries .

🧠 Conclusion: By integrating diverse drug modalities and leveraging data augmentation, Pisces inaugurates a new paradigm in computational pharmacology, offering a versatile and interpretable tool for discovering effective combination therapies and anticipating adverse interactions in clinical contexts.

📖 Reference:
Xu, H., Lin, J., Woicik, A., Liu, Z., Ma, J., Zhang, S., Poon, H., Wang, L., & Wang, S. (2025). Pisces: A multi-modal data augmentation approach for drug combination synergy prediction. Cell Genomics, 5, 100892. https://doi.org/10.1016/j.xgen.2025.100892

📜 License: This episode is based on an open-access article published under the Creative Commons Attribution 4.0 International License (CC BY 4.0).