贝叶斯优化的深度学习用于识别线粒体自噬的关键基因，并促进对抗人类癌症耐药性的疗法

Dysregulated mitophagy is essential for mitochondrial quality control within human cancers.

However, identifying hub genes regulating mitophagy and developing mitophagy-based treatments to combat drug resistance remains challenging.

Herein, BayeDEM (Bayesian-optimized Deep learning for identifying Essential genes of Mitophagy) was proposed for such a task. After Bayesian optimization, BayeDEM demonstrated its excellent performance in identifying critical genes regulating mitophagy of osteosarcoma (area under curve [AUC] of ROC: 98.

96%; AUC of PR curve: 100%). CERS1 was identified as the most essential gene regulating mitophagy (mean (|SHAP value|): 4. 14). Inhibition of CERS1 sensitized cisplatin-resistant osteosarcoma cells to cisplatin, restricting their growth, proliferation, invasion, migration and colony formation and inducing apoptosis.

Mechanistically, inhibition of CERS1 restricted mitophagy to destroy the mitochondrial quality control in cisplatin-resistant osteosarcoma cells, including mitochondrial membrane potential loss and unfavourable mitochondrial dynamics, rendering them susceptible to cisplatin-induced apoptosis. More importantly, mitophagy facilitated the immunosuppressive microenvironment formation by significantly modulating T-cell differentiation, adhesion and antigen presentation, and mitophagy mainly affects malignant osteoblasts in the early-mid developmental stage.

Immunologically, mitophagy potentially modulated the MIF signalling transmission between malignant osteoblasts and B cells, DCs, CD8+ T cells, NK cells and monocytes through the MIF-(CD74 + CXCR4) receptor-ligand interaction, thereby modulating the biological functions of these immune cells. Collectively, BayeDEM emerged as a promising tool for oncologists to identify pivotal genes governing mitophagy, thereby enabling mitophagy-centric therapeutic strategies to counteract drug resistance.