add support for multichain (residue_index) and real msa

[pengzhangzhi]

interesting pr. Would you like to elaborate on what you have done?

[sokrypton]

See more details here:

For multichain modeling (w/ residue index hack): https://twitter.com/sokrypton/status/1555114259209674752 For real msa: https://twitter.com/sokrypton/status/1555142289873313794

In short, we can hack OmegaFold to predict protein-protein interactions and replace the pseudo-MSA with real MSA :)

[sokrypton]

I'm not sure if the residue index is correctly implemented for the sin/cos RoPE module...

Still investigating.

[pengzhangzhi]

Terrific! Have you compare the result of a single sequence and multiple pseudo MSA? Does pseudo MSA helps?

[RuiWang1998]

Hi @pengzhangzhi,

Dr. @sokrypton here have already contacted us about the performance of MSA-enhanced OmegaFold. You can checkout his recent tweets about it here: https://twitter.com/sokrypton. In the mean time, thanks so much for your volunteering help here. We really appreciate it!

Dr. @sokrypton, Sorry for the delay, we are quite busy here at HeliXon currently. We are going to review the pull request and merge it (hopefully) in no time!

Thank you both so much for contributing!

Best, Rui

[sokrypton]

oops! looks like I accidentally deleted the branch while trying to merge latest commits from HeliXonProtein 😄

[RuiWang1998]

Hi Dr @sokrypton,

Sorry for the delay, but we have updated OmegaFold with significantly reduced memory overhead and would very much like to go on with this pull request if this is okay with you?

Best

[sokrypton]

Hi Rui, I'll work on merging the edits into GitHub. Seems lots of changes happened, that I'm no longer able to just do an automated merge 😅

On Sun, Aug 14, 2022, 12:50 PM Rui Wang @.***> wrote:

Hi Dr @sokrypton https://github.com/sokrypton,

Sorry for the delay, but we have updated OmegaFold with significantly reduced memory overhead and would very much like to go on with this pull request if this is okay with you?

Best

— Reply to this email directly, view it on GitHub https://github.com/HeliXonProtein/OmegaFold/pull/6#issuecomment-1214413816, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA76LASO5NPD6UJSM5WKDCLVZEPWRANCNFSM55TS4FIA . You are receiving this because you were mentioned.Message ID: @.***>

[RuiWang1998]

Hi Dr @sokrypton,

Great! We are sorry for the changes and delays. It is just that we think making the model accessible to more people is also very important.

Once you finish the edits, feel free to open a pull request!

Best

[sokrypton]

Sounds good!

On Tue, Jun 18, 2024, 8:24 AM Jack Maguire @.***> wrote:

@sokrypton https://github.com/sokrypton is this PR still on your radar? If not, would you be willing to proofread an attempt if I give it a shot?

— Reply to this email directly, view it on GitHub https://github.com/HeliXonProtein/OmegaFold/pull/6#issuecomment-2175974948, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA76LAVA2W66YGS6HMWGI2DZIARJTAVCNFSM6AAAAABJP5Q6KWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNZVHE3TIOJUHA . You are receiving this because you were mentioned.Message ID: @.***>