Hakken: Predicting future discoveries to fill gaps in today's knowledge (arxiv.org)

The recent announcement of Hakken, a novel prediction and explanation system, marks a significant advancement in the AI/ML community, particularly in scientific research. Hakken employs a transformer-based model that analyzes temporal sequences from knowledge graphs derived from extensive research papers. By leveraging a large language model's semantic understanding, it predicts unexplored relationships between scientific concepts, enhancing the growth of knowledge beyond existing deductions. Notably, Hakken has set a new standard for time-aware multi-label relation prediction in the biomedical field, demonstrating its versatility across various domains. In practical applications, Hakken scored 1.5 million hypotheses concerning aging, resulting in qualitative validations with biologists and three hypotheses advancing to wet-lab testing. Two of these predictions revealed previously unknown interactions—between TP53 and BAMBI, and RAF1 and TNF—holding substantial implications for drug discovery and repurposing. This capacity for identifying novel relationships not only fosters innovation in biomedical research but also exemplifies the potential of AI to fill critical knowledge gaps across disciplines.