\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16242374%2F2cf14a1eacb37c75d84fec54137d07e5%2FSnipaste_2025-03-27_16-11-54.png?generation=1743063136447997&alt=media\")
Acknowledgments: Special thanks to Yuki Kagaya, Daisuke Kihara, and the NuFold team
\nI've send some emails to them and here is what they shared. I've confirmed that I can share these details to the public as they are almost from the paper
Recycles: If you use the default parameter, the recycle would be 3. But actually results from the article are using 11±3 recycles.
\n\"I recommend setting recycling to around 15-20 and running NuFold to get the best result.\"
MSA: \"Each target has MSA files with the suffixes of base, meta, concat, and filter, all in A3M format. The CASP15 results were obtained by running inference four times using these MSAs individually.\" Here is the link for downloading MSA.
\nhttps://kiharalab.org/nufold/CASP15_MSA/
MSA generation: Using rMSA. The database versions used in the paper are either identical or slightly older than those mentioned in this discussion
Here are my results of NuFold on CASP15
\n
Acknowledgments: Special thanks to Yuki Kagaya, Daisuke Kihara, and the NuFold team
\nI've send some emails to them and here is what they shared. I've confirmed that I can share these details to the public as they are almost from the paper
Recycles: If you use the default parameter, the recycle would be 3. But actually results from the article are using 11±3 recycles.
\n\"I recommend setting recycling to around 15-20 and running NuFold to get the best result.\"
MSA: \"Each target has MSA files with the suffixes of base, meta, concat, and filter, all in A3M format. The CASP15 results were obtained by running inference four times using these MSAs individually.\" Here is the link for downloading MSA.
\nhttps://kiharalab.org/nufold/CASP15_MSA/
MSA generation: Using rMSA. The database versions used in the paper are either identical or slightly older than those mentioned in this discussion
Here are my results of NuFold on CASP15
\n
casp15 my proteinx results
\nhttps://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/566906#3161488
I'll probably trun to other af3 clones…
", "rawMarkdown": "I'll probably trun to other af3 clones..." } ] }, { "id": 3161431, "postDate": "2025-03-27T23:23:20.430Z", "content": "wondering how they created MSAs of self-distillation sequences where each MSA takes several hours…
", "rawMarkdown": "wondering how they created MSAs of self-distillation sequences where each MSA takes several hours...", "replies": [ { "id": 3161544, "postDate": "2025-03-28T04:45:19.633Z", "content": "i think MSA is limited by num of cpu thread (or num of gpu for dpu search).
\ni think these lab has PC cluster, and should not be a problem
Actually the results I shared above are not lb. They are from the validation set.
", "rawMarkdown": "Actually the results I shared above are not lb. They are from the validation set.", "votes": 1 } ] } ], "comments": [ { "id": 3161509, "author_name": "hengck23", "author_url": "", "post_date": "2025-03-28T03:38:48.263000", "content": "casp15 my proteinx results
\nhttps://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/566906#3161488
I'll probably trun to other af3 clones…
", "votes": 0, "replies": [] } ] }, { "id": 3161431, "author_name": "Zacchaeus", "author_url": "", "post_date": "2025-03-27T23:23:20.430000", "content": "wondering how they created MSAs of self-distillation sequences where each MSA takes several hours…
", "votes": 0, "replies": [ { "id": 3161544, "author_name": "hengck23", "author_url": "", "post_date": "2025-03-28T04:45:19.633000", "content": "i think MSA is limited by num of cpu thread (or num of gpu for dpu search).
\ni think these lab has PC cluster, and should not be a problem
Actually the results I shared above are not lb. They are from the validation set.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "3160853": "**Acknowledgments**: Special thanks to Yuki Kagaya, Daisuke Kihara, and the NuFold team\nI've send some emails to them and here is what they shared. I've confirmed that I can share these details to the public as they are almost from the paper\n\n- Recycles: If you use the default parameter, the recycle would be 3. But actually results from the article are using 11±3 recycles.\n\"I recommend setting recycling to around 15-20 and running NuFold to get the best result.\"\n\n- MSA: \"Each target has MSA files with the suffixes of base, meta, concat, and filter, all in A3M format. The CASP15 results were obtained by running inference four times using these MSAs individually.\" Here is the link for downloading MSA.\nhttps://kiharalab.org/nufold/CASP15_MSA/\n\n- MSA generation: Using rMSA. The database versions used in the paper are either identical or slightly older than those mentioned in [this discussion](https://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/569424)\n\nHere are my results of NuFold on CASP15\n", "3161509": "casp15 my proteinx results\nhttps://www.kaggle.com/competitions/stanford-rna-3d-folding/discussion/566906#3161488", "3161431": "wondering how they created MSAs of self-distillation sequences where each MSA takes several hours...", "3160866": "" } }