Big Data-Driven Computational Aptamer Design Framework via Parallel Monte Carlo Tree Search

Computational aptamer design using Monte Carlo Tree Search (MCTS) shows promise in drug discovery but is limited by sequential computing and simulation, which hinders the exploration of vast nucleotide sequence spaces and the integration of high-quality molecular docking techniques. We present a big data-driven parallel MCTS framework that significantly accelerates aptamer discovery while maintaining high accuracy. Our framework addresses these computational bottlenecks by distributing tasks across multiple processing units, enabling simultaneous exploration of diverse sequence regions and efficient incorporation of RNA/ssDNA-protein docking simulations for candidate evaluation. Key designs include a distributed search architecture with molecular docking, a score penalty system for optimized exploration, and comprehensive RNA/ssDNA docking pipelines that combine structure prediction with protein docking simulations. Experimental results targeting human complement component protein (i.e., protein 2A73) demonstrate a 98-fold increase in daily successful candidate discovery and improved binding affinity compared to the baseline, with the successful ratio increasing from 1% to 7.59%. Our approach not only accelerates the aptamer design process but also enhances the quality of generated aptamers, providing a more efficient and effective tool for drug discovery.