When AI meets AIE: Machine learning in design function-oriented AIEgens for antibacterial ability

Although photodynamic therapy ability of triphenylamine (TPA)-type aggregation-induced emission luminogens (AIEgens) for Gram-positive multidrug-resistant bacteria have been proofed, rational design of new function-oriented TPA-type AIEgens with high antibacterial ability is still limited by complicated experimental procedure and unknown antibacterial mechanism. Herein, we proposed one novel knowledge-based artificial intelligence (AI) approach combining theoretical calculation and machine learning to assist rationally design and predict antibacterial ability new TPA-type AIEgens. We firstly constructed one new TPA-type AIEgens abbreviated TBP-TA, pursuing lower energy gap and higher hydrophilicity that directly related to reactive oxygen species generation and bacteria binding ability, finally enhancing antibacterial ability. After confirmation of expected physicochemical features by theoretical calculation, the antibacterial ability of TBP-TA was predicted based on one homemade database containing 38 reported TPA-type AIEgens and machine learning models built in this work. As expected, TBP-TA exhibited antibacterial ability toward Gram-positive multidrug-resistant bacteria in prediction result provided by knowledge-based AI approach. Furthermore, in vitro and in vivo antibacterial results proved the excellent eliminating ability of TBP-TA to methicillin-resistant S. aureus comparable to vancomycin. Lastly, the antibacterial mechanism of TBP-TA was explored exhaustively. The knowledge-based AI approach we proposed represented a paradigm for the development of powerful AIE antibacterial agents with significantly reduced cost and increased success rate.