TusOAI: Agentic Optimization for Scientific Methods
1. TusOAI is an innovative AI system that autonomously develops and optimizes computational methods for scientific tasks. It integrates domain knowledge into a knowledge tree representation and performs iterative, domain-specific optimization and model diagnosis, significantly improving performance over existing methods.
2. The system outperformed state-of-the-art expert methods, MLE agents, and scientific AI agents across diverse tasks such as single-cell RNA-seq data denoising and satellite-based earth monitoring. This demonstrates its versatility and potential to accelerate scientific discovery.
3. TusOAI achieved a 40% power improvement to scDRS in associating cells to disease in simulations and a 10.5% enrichment improvement to pgBoost for identifying ground-truth variant-gene pairs. These enhancements highlight its ability to uncover novel biology.
4. The system also revealed 9 new associations between autoimmune diseases and T cell subtypes, as well as 7 previously unreported links between disease variants and their target genes. This showcases its potential to provide new biological insights.
5. TusOAI features a novel framework with a knowledge tree for structured representation of domain knowledge, hierarchical planning with Bayesian updates to balance solution quality and diversity, and fine-grained generation that integrates model optimization with diagnostic feedback.
6. The system was benchmarked on 6 single-cell analysis tasks and 5 scientific deep learning tasks, consistently outperforming baseline methods with an average 16% improvement in single-cell tasks. This highlights its robustness and effectiveness.
7. The code for TusOAI is publicly available, allowing researchers to explore and utilize this powerful tool for their own scientific endeavors.
📜Paper:
arxiv.org/abs/2509.23986
#AI #ScientificDiscovery #ComputationalMethods #Genetics #SingleCell #DeepLearning