Making AI-based scientific predictions more trustworthy

https://showme.missouri.edu/2026/making-ai-based-scientific-predictions-more-trustworthy/

Publish Date: 2026-02-18 08:40:00

Source Domain: showme.missouri.edu

Here are some key points from the article:

Release of PSBench Database: The University of Missouri has launched the world’s largest collection of protein models with quality assessments named PSBench, consisting of 1.4 million annotated protein structure models verified by independent experts.
Accelerating Drug Development: PSBench is anticipated to significantly accelerate drug development for diseases including Alzheimer’s and cancer by providing reliable information for building accurate AI systems to assess protein structure models.
Importance of Protein Structures: Proteins are fundamental to biological processes in the human body, and even small structural changes in their three-dimensional shapes can lead to diseases.
Advancements in AI: Recent advances in AI have greatly improved protein structure prediction; however, no single tool is consistently accurate across all types of proteins, necessitating a reliable benchmark like PSBench.
Benchmarking Tool: PSBench enables scientists to develop AI methods to evaluate the quality of predicted protein models and decide on their reliability, significantly advancing applications in disease understanding and treatment development.
Origin and Leadership: PSBench was developed by Jianlin “Jack” Cheng and his team at the University of Missouri, leveraging resources from the CASP competition, which is considered the gold standard for assessing computational methods for protein prediction.
Deep Learning Contribution: Cheng’s earlier work in deep learning demonstrated its effectiveness in solving protein folding problems, leading to transformative advancements, including the development of AlphaFold.
Global Impact: Presented at NeurIPS 2025, a prestigious AI conference, PSBench is seen as a major contribution to the global scientific community, pushing forward AI-driven biomedical discoveries.