Interpretable neural networks prioritize cancer driver genes from genome-wide dependency landscapes.

Interpretable neural networks identify cancer driver genes from genome-wide dependency landscapes, recovering major drivers and highlighting candidates with broad functional footprints.

• Why it matters: Accurate identification of cancer driver genes and their therapeutic relevance remains a key challenge in computational cancer biology, requiring methods that can interpret complex genomic data.
• What they did: The study introduces two neural network frameworks, xNNDriver and xAEDriver, analyzing 3,008 candidate genes by connecting mutation status with gene dependencies, pathway activity, and drug responses through supervised and unsupervised models.
• The result: These models demonstrate that gene dependency landscapes encode interpretable signals of oncogenic function, enabling prioritization of drivers, pathways, and vulnerabilities for experimental validation.