Have you heard about the case of the missing HIV drug, ritonavir? In the late 90’s, Abbott Labs suddenly lost the ability to purify this life-saving drug, and intensive efforts were made by many chemists to find the cause. Turns out, the culprit was rampant formation of the compound into an undesired crystal form with no bioactivity, referred to as crystal polymorphism. Now, the pharmaceutical industry knows the importance of understanding and mitigating risks of unexpected crystal polymorphs in small molecule drugs, which can have vastly different physical and chemical properties. What if it was possible to predict novel crystal structure prediction for molecular crystals and avoid such quality, efficacy, and safety issues? Dong Zhou recently authored a paper, out this month in Nature Portfolio #NatureCommunications, answering this precise question. At Schrodinger, he and his co-authors developed a novel crystal structure prediction (CSP) method for molecular crystals, offering both exceptional accuracy and efficiency. The method features a systematic divide-and-conquer crystal packing search algorithm, built upon space group symmetry decompositions, as well as machine learning force fields for hierarchical energy ranking. This high accuracy and reliability enables routine crystal structure prediction in drug formulation, accelerating clinical formulation design and de-risking downstream manufacturing processes. Learn more by checking out the article here: https://lnkd.in/e7PrXBSb. At Atommap, Dong Zhou uses his molecular physics expertise to develop a next-generation force field that combines first-principle quantum calculations with modern machine learning. This enables us to enhance accuracy while maintaining computational efficiency in small molecule modeling. #AIDrugDiscovery #MachineLearning #DrugDiscovery #DrugDevelopment #Quantum Check out this cool PBS video on ritonavir: https://lnkd.in/eJN3zqx8
关于我们
Atommap invents therapeutic molecules using quantum physics, AI, and computation-driven design at atomic precision.
- 网站
-
https://atommap.com
Atommap的外部链接
- 所属行业
- 生物技术研究
- 规模
- 2-10 人
- 总部
- New York，New York
- 类型
- 私人持股
- 创立
- 2023
- 领域
- Computational Chemistry、Machine Learning、Molecular Design和Drug Discovery
地点
-
主要
获取路线
450 Lexington Avenue
US，New York，New York，10017
Atommap员工
动态
-
Our takeaway from last week’s Next in TechBio event in sunny San Francisco at #JPM2025: AI-driven drug discovery needs to solve well-defined and critical problems in drug discovery, and it will take patience and focus. We are grateful to the community of scientists, investors, and dealmakers that gathered for the discussion among our CEO Huafeng Xu, Felix Wong, Mark DePristo, Thomas Clozel, Robert Marino and Thomas Fuchs on AI in drug discovery & development across different modalities. Special thanks to Integrated Biosciences, Inc. for organizing, Benchling for use of their beautiful space, and SVB, Wilson Sonsini Goodrich & Rosati, and Illumina Ventures for sponsoring.
-
-
What does the future hold for techbio??What role will AI and computation play in drug discovery??On Tuesday 1/14, Huafeng Xu will be exploring these topics with a group of leaders at the #JPM2025 Next in TechBio cocktail event hosted by Integrated Biosciences, Inc. and sponsored by Benchling, Wilson Sonsini Goodrich & Rosati, and SVB RSVP: https://lnkd.in/ehY-PE3t
-
-
Seldom does one coauthor a paper with a college classmate from decades ago. I am happy to share one such paper, with Weiping Tang, who was my classmate and dorm neighbor in Peking University, and others. https://lnkd.in/eZftFJqE In this paper, we explore how advanced molecular simulations may help design targeted protein degraders. Degrader molecules bring together a E3 ligase and a target protein to form a ternary complex, which facilitates ubiquitin transfer to the target and its subsequent degradation. There are often many ways that the ligase and the target protein can come together. Predicting the most favorable protein-protein interfaces may inform the design of degraders that induce stable ternary complexes and efficient ubiquitin transfer. Although not reported in this paper, we have computationally designed--and subsequently validated by wet-lab assays--potent degrader molecules based on such modeled ligase-protein complexes. By a similar approach we have recently identified molecular glues that degrade an oncogenic mutant protein, for which no molecular glue degraders have been reported. These are still early days of computation-driven design of targeted protein degraders. I look forward to more fun working together with old friends and new collaborators.