predictioncenter.org

One-line Introduction

predictioncenter.org is a global protein structure prediction assessment platform funded by organizations such as the U.S. National Institutes of Health (NIH) and operated within the CASP (Critical Assessment of Structure Prediction) competition framework. Researchers can submit prediction models for free and compare them fairly against peers worldwide, helping advance structural biology and computational biology. It is not a commercial tool, but rather the academic community’s recognized “referee” and benchmarking platform.

Business Overview

The platform’s core activity is organizing the biennial CASP competition—the most authoritative blind assessment in the global protein structure prediction field. Participating teams from universities, research institutes, and corporate labs predict protein structures that have not yet been publicly solved, after which the CASP organizers compare and score the submissions against experimentally determined structures. The platform also provides historical datasets, evaluation metrics such as GDT_TS and LDDT, and leaderboards for researchers to study and cite. Since its launch in 1994, CASP has become the standard stage for evaluating leading methods such as AlphaFold. Its main users are structural biologists, computational chemists, and AI drug discovery teams, who care about objective validation of methods rather than commercial monetization.

Who It’s For

• Academic researchers: For validating prediction algorithms or deep learning models and comparing them against benchmark methods such as AlphaFold and RoseTTAFold under a unified standard.
• Drug development teams: For using blind assessments to verify the reliability of in-house protein structure prediction tools in small-molecule docking and antibody design.
• Bioinformatics students: For learning the principles of structure prediction and using CASP historical datasets in coursework or thesis experiments.
• Not suitable for: Purely industrial use cases, such as directly obtaining commercial-grade structure prediction results, or non-research users. The platform does not provide prediction services; it only provides assessment.

Key Features and Highlights

• Blind-test competition mechanism: Held every two years; target protein structures are kept confidential by experimental groups before release, preventing prediction cheating.
• Standardized evaluation metrics: Authoritative scoring systems such as GDT_TS, LDDT, and RMSD, with reproducible and citable results.
• Open historical database: Submission results, rankings, and target structures from all previous CASP editions can be downloaded for free, with registration required.
• Recognized by the global community: CASP rankings are an important reference for top journals such as Nature and Science, as well as grant review panels.
• Free participation: After registering an account, any researcher can submit predictions for free, with no hidden fees.
• Deeply associated with AlphaFold: DeepMind showcased AlphaFold at CASP13 and CASP14, helping establish it as an industry benchmark.

Pricing Analysis

Completely free. This is the platform’s most important advantage—it does not charge users for submitting predictions, downloading data, or viewing rankings. However, users must cover their own computing resource costs, such as GPU servers and cloud platform fees, as well as any potential data transfer costs. Compared with commercial tools such as Schrödinger’s Prime or paid Rosetta licenses, predictioncenter.org costs nothing to participate in. That said, it does not offer paid acceleration or customer support; all services are self-service under community rules.

How Chinese Users Can Use It

• Network accessibility: Direct access from mainland China is generally smooth, with no VPN required. predictioncenter.org and its subdomains, such as predictioncenter.org/casp14, are not blocked, and speeds are moderate.
• Payment methods: No payment is required. Registration only needs an email address, preferably an edu.cn or institutional email. No credit card or third-party payment binding is needed.
• Is a VPN/proxy needed?: No. However, when downloading large datasets, such as historical prediction PDB files, it is recommended to use a campus network or stable broadband connection, as some files may be large.
• Domestic alternatives: There is currently no fully comparable free platform in China. Similar domestic projects include the China Protein Structure Prediction Competition (CPC) or deep-learning protein structure prediction platforms such as PaddleHelix, but their influence, dataset scale, and international recognition are far below CASP.
• Invoices: Since the platform is free, it does not issue invoices. If you need reimbursement for CASP-related travel or computing costs, you will need to handle this through your institution’s finance department.

Pros and Cons

Pros:

• ✅ Completely free, with no hidden fees
• ✅ The world’s most authoritative structure prediction assessment standard
• ✅ Rich historical data, suitable for research reproducibility and teaching
• ✅ Direct access from mainland China, no VPN required
• ✅ Active community and high citation value in academic papers

Cons:

• ❌ Does not provide prediction services; it only performs assessment, so users must develop or run their own models
• ❌ Long competition cycle, held once every two years, making it unsuitable for rapid validation
• ❌ No Chinese interface or Chinese-language support
• ❌ No refund policy, since it is free, but also no service guarantee
• ❌ Large dataset downloads may occasionally experience packet loss on domestic networks

Comparison with Similar Products

• AlphaFold DB: A protein structure database maintained by Google/DeepMind that provides free prediction results, but does not offer third-party blind assessment. predictioncenter.org is the “referee,” while AlphaFold DB is a “contestant.”
• Rosetta Commons: An open-source protein modeling suite that provides prediction tools, but users must organize assessment themselves. predictioncenter.org provides standardized assessment, while Rosetta focuses more on algorithm development.
• I-TASSER: A free protein structure prediction server developed by the University of Michigan that directly outputs models, but does not use a unified assessment standard. predictioncenter.org only performs assessment and does not output models.

Summary and Recommendation

Recommended if: You are a researcher in structure prediction and need to prove that your algorithm is better than or comparable to current leading methods, or you want high-confidence benchmarking results for publication. In that case, predictioncenter.org is the only free option with this level of authority. It is recommended to register an account first, download the CASP14 or CASP15 datasets for local testing, and become familiar with the submission process before entering the official competition.

Not recommended if: You simply want to quickly obtain a structural model for a specific protein, such as for drug screening. In that case, use AlphaFold DB, I-TASSER, or RoseTTAFold directly. If you need Chinese-language support or commercial-grade service, this platform will not meet your needs.

Suggested action: It is completely free, so there is little reason to hesitate. Visit the official website, register an account, and join the mailing list to receive notifications about the next CASP competition.