【Nat.Rev.Mater.】新加坡国立刘斌|高性能有机光敏剂分子设计的3大机制与机器学习自驱动发现前沿综述#

The main photosensitization pathways are classified into three different groups: energy-transfer-based type II photosensitizers (PSs, top panel), electron-transfer-based type I PSs (left panel) and oxygen-independent PSs through direct substrate activation (right panel). Bottom panel: schematic machine-learning (ML) workflow for PS discovery. ISC, intersystem crossing; PS*, excited-state PS; S0, singlet ground state; S1, lowest singlet excited state; T1, lowest triplet excited state.#

a, Typical molecular structures of type II photosensitizers (PSs). b, Schematic illustration of highest occupied molecular orbital (HOMO)–lowest unoccupied molecular orbital (LUMO) engineering for fine-tuned Δ_E_ST (singlet–triplet energy gap). c, Emerging strategies for facilitating the intersystem crossing (ISC) process: schematic illustration of ISC engineering via enhancing spin–orbit coupling (SOC) or reducing Δ_E_ST; heteroatom substitution; twisted molecular conformations; and aggregation-enhanced ISC through self-assembly of chromophores. d, Molecular engineering of aggregation-induced emission (AIE) PSs through D–A engineering, π-bridge extension and polymerization. e, Absorption modulation through the Förster resonance energy transfer (FRET) mechanism. A, acceptor; D, donor; S0, singlet ground state; S1, lowest singlet excited state; T1, lowest triplet excited state; TPE, tetraphenylethylene.#

第二类是对氧气依赖性较低的I型机制，它通过激发态光敏剂与周围底物发生一系列电子转移，生成超氧阴离子自由基、羟基自由基和过氧化氢。由于I型机制需要激发态能量有效竞争过能量转移路径，其设计难度通常更高。目前主要的构筑策略包括能量转移阻断，即通过合理控制光敏剂的三线态能级，使其低于敏化氧气产生单线态氧的阈值，从而选择性地触发I型光路。同时，利用外围结构修饰进行电子转移编程，引入强吸电子或供电子基团作为“电子泵”或“电子水库”，能有效重塑激发态的电子流向。此外，通过聚集态下的同质分子自组装诱导对称性破缺电荷分离，或者在单分子内共价连接特异性电子供受体构建光诱导电荷分离体系，能够稳定生成自由基离子对，从而极大地加速I型活性氧物种的释放。

a, Typical structures of type I photosensitizers (PSs). b, Energy-transfer blocking strategy, in which synergistic intersystem crossing (ISC) enhancement and suppression of the energy-transfer pathway selectively facilitate type I reactive oxygen species (ROS) generation. c, Electron-transfer programming strategy, in which covalent incorporation of electron-donating or electron-withdrawing units directs excited-state electron flow towards type I ROS generation. d, Photoinduced charge separation strategy, in which intermolecular (top panel) or intramolecular (bottom panel) electron transfer generates radical pairs that drive type I ROS formation. D–A, donor–acceptor; ES, excited state; ET, electron transfer; GS, ground state; 1PS*, singlet excited state; 3PS*, triplet excited state; PS•−, photosensitizer anion radical; PS•+, photosensitizer cation radical; SOC, spin–orbit coupling; S0, ground state; S1, lowest singlet excited state; T1, lowest triplet excited state; Δ_E_ST, singlet–triplet energy gap.#

第三类则是新兴的无氧激发机制，这类光敏剂能够在严重缺氧的环境中直接激活周围的底物，完美绕过了对外界分子氧的依赖，为解决生物体内肿瘤缺氧耐受性等极端环境应用提供了创新的分子解法。无氧光敏剂可细分为氧自给型和氧旁路型两种设计思路。氧自给型光敏剂主要通过模拟水氧化机制，利用光生空穴的强氧化电位将水直接氧化为氧气和氢离子，生成的氧气随即作为中间体在微环境中参与后续的活性氧转化，从而实现缺氧环境下的自循环光催化。氧旁路型光敏剂则通过精准的轨道能量匹配与空间紧密结合，将激发态的能量、电子或空穴直接输送给目标生物大分子或底物，引发特定化学键断裂或还原反应，实现高选择性的直接底物激活。此外，分子设计还可以扩展到多组分复合物体系，通过引入外部电子供体如血清白蛋白或电子受体如百里醌，动态调节电子和能量的分配格局，从而在不同活性氧路径之间实现智能的可控转换。

Typical thermodynamic (panel a) and kinetic (panel b) characteristics and reaction mechanisms of oxygen-independent photosensitizers (PSs), encompassing both oxygen self-sufficient and oxygen-bypassed PSs. Thermodynamic and kinetic design strategies of oxygen self-sufficient (panel c) and oxygen-bypassed (panel d) PSs. PS*, excited-state photosensitizer; RedPS, reduction potential of PS; RedSubstrates, reduction potential of substrates; S1, lowest singlet excited state; T1, lowest triplet excited state.#

a–d, Schematic illustration of electron-transfer-based photosensitization pathway modulation, including electron donor-mediated photoinduced electron transfer (PET, panel a) and electron acceptor-mediated PET (panel c). Representative structures of electron donors (panel b) and electron acceptors (panel d) are presented. e,f, Schematic illustration of energy-transfer-based photosensitization modulation through FRET (panel e) and triplet–triplet energy transfer (panel f). A, acceptor; D, donor; ISC, intersystem crossing; PS, photosensitizer; TA, tertiary amine; TIPS, triisopropylsilyl.#

为了彻底攻克光敏剂结构-性能关系难以捉摸的难关，机器学习（ML）正在成为加速新型分子发现的关键底座。在监督学习框架下，研究者利用分子指纹、前线轨道能量以及量子化学描述符，成功构建了能够高效预测氟硼荧（BODIPY）、卟啉和花菁等主流光敏剂家族量子产率与生物活性的分类与回归模型，并在百万级的大规模分子库中筛选出了性能优异的候选分子。进一步地，主动学习将图卷积网络代理模型与密度泛函理论计算紧密耦合，形成闭环反馈体系，通过贝叶斯优化在庞大的分子空间中迭代选择最具信息量的数据进行再训练，极大地提高了高性能分子的命中率。在此基础上，深度生成模型引入强化学习算法，能够从断裂的化学片段或SMILES字符串中从头构筑满足吸收波长与能量激发双重约束的全新分子骨架，彻底打破了传统合成库的边界限制。

a, Supervised learning models use molecular representations with statistical or deep-learning algorithms. b, Active learning integrates machine learning (ML) models with density functional theory (DFT) and time-dependent DFT (TD-DFT) calculations in a feedback loop, in which Bayesian optimization suggests new candidates for evaluation and retraining. c, Generative pipelines apply ML models with reinforcement learning to propose novel scaffolds. d, Representative and experimentally validated molecular structures obtained from different ML approaches. PS, photosensitizer.#