AI/ML时代的精准噬菌体治疗：从发现到临床转化证据的系统综述

In precision phage therapy, artificial intelligence and machine learning (AI/ML) is being applied for phage detection, genomes/proteome annotation, host prediction, resistance profiling and therapeutic optimization. Yet the landscape remains fragmented owing to a lack of systematic synthesis of its translational maturity, methodological robustness and pipeline coverage. We performed a systematic review of AI/ML tools relevant to precision phage therapy. Web of Science, PubMed, MEDLINE, Scopus and IEEE Xplore were used to identify records. Of 6,969 identified records, 2,214 duplicates were removed, leaving 4,755 unique records for title/abstract screening; 510 full-text articles were assessed for eligibility, and 128 studies were included in the final synthesis. Studies were classified across a nine-module precision phage therapy pipeline by architectural family, validation maturity, accessibility, translational readiness (AI_PTRL), and methodological quality using an adapted PROBAST framework. The 128 included studies were published between 2012 and 2025, with a median publication year of 2023 and peak activity during 2022-2025. The large majority were methodological/tool development studies (115/128, 89.8%) and these predominantly described in silico validation (101/128, 78.9%). The pipeline coverage was uneven, with the highest concentration in the host range and interaction determinants (M4, 43/128, 33.6%) and genome/protein annotation and functional inference (M3, 33/128, 25.8%). Safety screening, therapeutic design, and infrastructure modules were comparatively underrepresented. From an architectural standpoint, hybrid/integrated systems (HIS) (39/128, 30.5%) and classical machine learning (CML) (38/128, 29.7%) were the most common approaches with substantial diversification after 2022. The overall translational readiness was modest (median AI_PTRL: 6, IQR 4-6). Methodological assessment demonstrated high evaluation risk of bias (104/128, 81.2%) and substantial development concern (88/128, 68.8%), especially in Domain 4 (analysis). AI/ML for precision phage therapy is a rapidly growing and diversifying field, particularly in host-phage interaction prediction, however efforts are still limited for downstream precise phage therapy tasks. Experimental validation and real-world deployment remain limited with methodological limitations due to analytical design and evaluative approaches. Future work should prioritize rigorous validation frameworks, higher-quality and more complete representative datasets, stronger safety screening, therapeutic optimization, and clinically actionable decision support.