The 17th Critical Assessment for Structure Prediction (CASP17) has begun! This is the world's flagship challenge for predictive biology, and it was AlphaFold's performance in CASP14 that led to a Nobel prize in 2024 (for protein structure prediction).
\nExcitingly, all three top Kaggle teams in this RNA modeling competition (team_cp, AyPy, and Stefan Stefanov) are collaborating to make their algorithms into automated servers (see the CASP17 server status page).
\nThis effort has required updating codes to produce all-atom predictions and to handle the CASP PDB format. We’re grateful for the collaboration.
\nNow if you are also interested in participating, it is not too late.
\nThere are a specific set of targets labeled for human prediction with an ‘h’ in the target Type on the CASP17 RNA target list. The current set available for prediction are:
\nAnd there will be more. These and other ‘human’ prediction targets will likely be assessed together, separately from several ‘server-only’ targets that have completed already.
\nFor these targets, you don’t have to upload any code and you don’t even have to set up a CASP17 server. You are welcome to just use your existing Kaggle notebooks to generate models that you can then submit ‘by hand’.
\nPlease note that CASP requires all atoms so you may want to retain those from your code, or, as a backstop, impute them from C1' only models in with Arena. To prevent gaming metrics, the assessors will choose or develop a range of metrics after receiving the predictions, so consider looking at metrics like lDDT or RMSD in addition to this Kaggle competition's TM-score.
\nThe first deadline for RNA human targets is June 01, 2026 and several more RNA targets will be coming through August, 2026. If you start submitting with the first target R2341, you will avoid any penalties for missing predictions. Final results will then be announced in December, 2026.
\nIf you’re interested in seeing how your models do compared to the leading groups in 2026, register here!
\nFeel free to post here with questions or tips.
\nFor the hosts,\nRhiju @rhijudas\nPrzemek @przemekporebski
\nP.S. Some of the CASP17 server-only targets overlap with the test set in this Kaggle Stanford RNA 3D Folding Challenge Part 2. For that reason, we will re-open this Kaggle competition for late submissions in August, when CASP17 closes. You can get an ongoing preview of results by looking for the experimental structures of CASP17 server-only targets as they come out in the PDB every Wednesday, e.g., for the four AI-designed RNA molecules from this preprint.
", "messages": [ { "id": 3463408, "postDate": "2026-05-26T20:14:13.660Z", "content": "The 17th Critical Assessment for Structure Prediction (CASP17) has begun! This is the world's flagship challenge for predictive biology, and it was AlphaFold's performance in CASP14 that led to a Nobel prize in 2024 (for protein structure prediction).
\nExcitingly, all three top Kaggle teams in this RNA modeling competition (team_cp, AyPy, and Stefan Stefanov) are collaborating to make their algorithms into automated servers (see the CASP17 server status page).
\nThis effort has required updating codes to produce all-atom predictions and to handle the CASP PDB format. We’re grateful for the collaboration.
\nNow if you are also interested in participating, it is not too late.
\nThere are a specific set of targets labeled for human prediction with an ‘h’ in the target Type on the CASP17 RNA target list. The current set available for prediction are:
\nAnd there will be more. These and other ‘human’ prediction targets will likely be assessed together, separately from several ‘server-only’ targets that have completed already.
\nFor these targets, you don’t have to upload any code and you don’t even have to set up a CASP17 server. You are welcome to just use your existing Kaggle notebooks to generate models that you can then submit ‘by hand’.
\nPlease note that CASP requires all atoms so you may want to retain those from your code, or, as a backstop, impute them from C1' only models in with Arena. To prevent gaming metrics, the assessors will choose or develop a range of metrics after receiving the predictions, so consider looking at metrics like lDDT or RMSD in addition to this Kaggle competition's TM-score.
\nThe first deadline for RNA human targets is June 01, 2026 and several more RNA targets will be coming through August, 2026. If you start submitting with the first target R2341, you will avoid any penalties for missing predictions. Final results will then be announced in December, 2026.
\nIf you’re interested in seeing how your models do compared to the leading groups in 2026, register here!
\nFeel free to post here with questions or tips.
\nFor the hosts,\nRhiju @rhijudas\nPrzemek @przemekporebski
\nP.S. Some of the CASP17 server-only targets overlap with the test set in this Kaggle Stanford RNA 3D Folding Challenge Part 2. For that reason, we will re-open this Kaggle competition for late submissions in August, when CASP17 closes. You can get an ongoing preview of results by looking for the experimental structures of CASP17 server-only targets as they come out in the PDB every Wednesday, e.g., for the four AI-designed RNA molecules from this preprint.
", "rawMarkdown": "The 17th Critical Assessment for Structure Prediction (CASP17) has begun! This is the world's flagship challenge for predictive biology, and it was AlphaFold's performance in CASP14 that led to a Nobel prize in 2024 (for protein structure prediction).\n\nExcitingly, all three top Kaggle teams in this RNA modeling competition (**team_cp**, **AyPy**, and **Stefan Stefanov**) are collaborating to make their algorithms into automated servers (see the [CASP17 server status page](https://predictioncenter.org/casp17/status_n.cgi)). \n\nThis effort has required updating codes to produce all-atom predictions and to handle the CASP PDB format. We’re grateful for the collaboration.\n\nNow if **you** are also interested in participating, it is not too late.\n\nThere are a specific set of targets labeled for human prediction with an ‘h’ in the target Type on the [CASP17 RNA target list](https://predictioncenter.org/casp17/targetlist.cgi?view=rna ). The current set available for prediction are:\n\n- [R2341](https://predictioncenter.org/casp17/target.cgi?id=76&view=rna) - RRE core\n- [R2350](https://predictioncenter.org/casp17/target.cgi?id=90&view=rna) - Tetrahymena Ribozyme C7.2 mutant RNA \n- [R2351](https://predictioncenter.org/casp17/target.cgi?id=91&view=rna) - Tetrahymena Ribozyme R1 mutant RNA\n\nAnd there will be more. These and other ‘human’ prediction targets will likely be assessed together, separately from several ‘server-only’ targets that have completed already. \n\nFor these targets, you don’t have to upload any code and you don’t even have to set up a CASP17 server. You are welcome to just use your existing Kaggle notebooks to generate models that you can then submit ‘by hand’. \n\nPlease note that CASP requires all atoms so you may want to retain those from your code, or, as a backstop, impute them from C1' only models in with Arena. To prevent gaming metrics, the assessors will choose or develop a range of metrics after receiving the predictions, so consider looking at metrics like lDDT or RMSD in addition to this Kaggle competition's TM-score.\n\nThe first deadline for RNA human targets is **June 01, 2026** and several more RNA targets will be coming through August, 2026. If you start submitting with the first target R2341, you will avoid any penalties for missing predictions. Final results will then be announced in December, 2026. \n\nIf you’re interested in seeing how your models do compared to the leading groups in 2026, register [here](https://predictioncenter.org/casp17/registration.cgi)!\n\nFeel free to post here with questions or tips.\n\nFor the hosts,\nRhiju @rhijudas\nPrzemek @przemekporebski \n\n**P.S.** Some of the CASP17 server-only targets overlap with the test set in this [Kaggle Stanford RNA 3D Folding Challenge Part 2](https://www.kaggle.com/c/stanford-rna-3d-folding-2/). For that reason, we will re-open this Kaggle competition for late submissions in August, when CASP17 closes. You can get an ongoing preview of results by looking for the experimental structures of CASP17 server-only targets as they come out in the PDB every Wednesday, e.g., for the four AI-designed RNA molecules from [this preprint](https://www.biorxiv.org/content/10.64898/2026.05.21.726960v1).\n", "votes": 2 } ], "comments": [], "raw_markdown_by_id": { "3463408": "The 17th Critical Assessment for Structure Prediction (CASP17) has begun! This is the world's flagship challenge for predictive biology, and it was AlphaFold's performance in CASP14 that led to a Nobel prize in 2024 (for protein structure prediction).\n\nExcitingly, all three top Kaggle teams in this RNA modeling competition (**team_cp**, **AyPy**, and **Stefan Stefanov**) are collaborating to make their algorithms into automated servers (see the [CASP17 server status page](https://predictioncenter.org/casp17/status_n.cgi)). \n\nThis effort has required updating codes to produce all-atom predictions and to handle the CASP PDB format. We’re grateful for the collaboration.\n\nNow if **you** are also interested in participating, it is not too late.\n\nThere are a specific set of targets labeled for human prediction with an ‘h’ in the target Type on the [CASP17 RNA target list](https://predictioncenter.org/casp17/targetlist.cgi?view=rna ). The current set available for prediction are:\n\n- [R2341](https://predictioncenter.org/casp17/target.cgi?id=76&view=rna) - RRE core\n- [R2350](https://predictioncenter.org/casp17/target.cgi?id=90&view=rna) - Tetrahymena Ribozyme C7.2 mutant RNA \n- [R2351](https://predictioncenter.org/casp17/target.cgi?id=91&view=rna) - Tetrahymena Ribozyme R1 mutant RNA\n\nAnd there will be more. These and other ‘human’ prediction targets will likely be assessed together, separately from several ‘server-only’ targets that have completed already. \n\nFor these targets, you don’t have to upload any code and you don’t even have to set up a CASP17 server. You are welcome to just use your existing Kaggle notebooks to generate models that you can then submit ‘by hand’. \n\nPlease note that CASP requires all atoms so you may want to retain those from your code, or, as a backstop, impute them from C1' only models in with Arena. To prevent gaming metrics, the assessors will choose or develop a range of metrics after receiving the predictions, so consider looking at metrics like lDDT or RMSD in addition to this Kaggle competition's TM-score.\n\nThe first deadline for RNA human targets is **June 01, 2026** and several more RNA targets will be coming through August, 2026. If you start submitting with the first target R2341, you will avoid any penalties for missing predictions. Final results will then be announced in December, 2026. \n\nIf you’re interested in seeing how your models do compared to the leading groups in 2026, register [here](https://predictioncenter.org/casp17/registration.cgi)!\n\nFeel free to post here with questions or tips.\n\nFor the hosts,\nRhiju @rhijudas\nPrzemek @przemekporebski \n\n**P.S.** Some of the CASP17 server-only targets overlap with the test set in this [Kaggle Stanford RNA 3D Folding Challenge Part 2](https://www.kaggle.com/c/stanford-rna-3d-folding-2/). For that reason, we will re-open this Kaggle competition for late submissions in August, when CASP17 closes. You can get an ongoing preview of results by looking for the experimental structures of CASP17 server-only targets as they come out in the PDB every Wednesday, e.g., for the four AI-designed RNA molecules from [this preprint](https://www.biorxiv.org/content/10.64898/2026.05.21.726960v1).\n" } }