Zhanglab forum

Dear DMFold authors,

Thank you very much for the very useful tool. I am trying to use DeepMSA output as the input into ColabFold, but ColabFold needs MSA for both paired sequences and the two sequences alone. May I ask how I can derive the desired format from DeepMSA output? If DeepMSA can work with AlphaFold, I am also happy to learn how to put DeepMSA output into AlphaFold. Meanwhile, if DMFold itself can accept custom MSAs and templates, I am also eager to learn how to use it because I have many similar sequences - whose MSAs and templates can be the same - to predict structure from. Any help or guidance is highly appreciated!

Hi max64lowes,

Thanks for your interest in our work!

Regarding your first question — how to obtain the desired (per-chain) format from DeepMSA2 multimer output

DeepMSA2 produces paired MSAs for multimers (all chains concatenated into one alignment). To convert this into unpaired, per-chain A3M files, split each aligned sequence in the paired A3M according to the query chain order and their query lengths.

Procedure

• Note the chain order and query lengths from your input FASTA (e.g., A, B, C with lengths L_A, L_B, L_C).
• For each sequence line in the paired A3M:
  • Use the aligned query lengths as cut points: take the first L_A columns as chain A, the next L_B as chain B, and the next L_C as chain C.
• Write the extracted blocks into separate A3M files (one per chain).
  
  Example
  If chains A, B, C have lengths 10, 10, and 20, respectively, then each paired A3M sequence is split into:
  • columns 1–10 → A3M(A)
  • columns 11–20 → A3M(B)
  • columns 21–40 → A3M(C)

Regarding the second question — combining DeepMSA with AlphaFold

In DeepMSA2, DMFold refers to the integrated workflow of DeepMSA2 and AlphaFold: MSAs generated by DeepMSA2 are used in place of AlphaFold’s default MSA search step, while the rest of the AlphaFold pipeline (template search, model inference, relaxation) remains unchanged. If you wish to use this combination, simply run DMFold—it already performs this substitution and feeds the DeepMSA2 A3M files directly into AlphaFold.

Regarding the third question — supplying custom MSAs and templates to DMFold

The current DMFold server build does not support custom MSAs or templates. However, you can download the DMFold source and inspect MSA_combination.py to locate the MSA output directory; placing your custom MSA in that path will work. Note that template customization is not supported in the present DMFold release. Alternatively, you may download the AlphaFold2 source and modify it to read your own MSAs and templates.