Ecosystem is exposed to complex mixtures. Identifying causative toxicants is a key for understanding mixture risks but there are many data gaps among chemical and bioactivity data. In the big data era, scientific decision based on data mining and machine learning has been proposed as a powerful strategy. The collection retrospective risk-related big data can greatly promote scientific decisions on endpoint selection rather than traditional arbitrary decision-making that merely depends on expert judgment. We proposed an event-driven taxonomy (EDT) concept to integrate adverse outcome pathway (AOP)-based meta-data in risk assessments. Using EDT, molecular initiating events and the substances causing the events were fused in a data matrix, which was named an event driver (ED). From aquatic toxicity assessments within China over the past decade, we gathered over 14,000 sources of information. With a dictionary that included 1039 toxicological terms, 15 toxic modes of actions were mapped, yet less than half of the bioassays could be elucidated by available AOPs. A Naïve Bayesian ED-classifier was developed to mechanistically annotate apical responses. The classifier reached 74% accuracy and labeled 85% bioassays as 26 EDs. Narcosis, estrogen receptor- and aryl hydrogen receptor-mediators were the major EDs in aquatic systems across China. Individual regions had distinct ED fingerprints. Subsequently, an EDT-based artificial intelligence-assisted integrated testing strategy (ITS) was constructed for aquatic mixture risk assessment by integrating high-throughput screening and chemical predictions. This EDT-based ITS was evaluated using complex sediment mixtures eliciting arylhydrocarbon receptor activation and oxidative stress response. While mixture prediction using expert knowledge-oriented target analysis hardly explained sediment toxicity, a metadata-driven suspect analysis explained >80% toxicity. Furthermore, deep learning models were developed to extract fingerprints of bioactive suspect candidates and convert these fingerprints to mass spectra-readable information for non-target screening with GC-qToF-MS. The EDT-based ITS tool provides a promising strategy for mixture risk assessments in a big data era.